From 8790537917e7d57eaf03f54733e0ddd50d0a3b91 Mon Sep 17 00:00:00 2001 From: Teriks Date: Sun, 20 Apr 2025 21:52:19 -0500 Subject: [PATCH 1/3] Kolors additional pipelines, community contrib Adapted from: https://github.com/Kwai-Kolors/Kolors Mostly for direct use with diffusers existing ControlNetModel code + an additional pipeline for inpainting. Pipelines here: KolorsControlNetPipeline KolorsControlNetImg2ImgPipeline KolorsControlNetInpaintPipeline KolorsInpaintPipeline Complete doc, but could use a look over. --- .../pipeline_controlnet_xl_kolors.py | 1306 ++++++++++++ .../pipeline_controlnet_xl_kolors_img2img.py | 1443 +++++++++++++ .../pipeline_controlnet_xl_kolors_inpaint.py | 1807 +++++++++++++++++ .../community/pipeline_kolors_inpainting.py | 1726 ++++++++++++++++ 4 files changed, 6282 insertions(+) create mode 100644 examples/community/pipeline_controlnet_xl_kolors.py create mode 100644 examples/community/pipeline_controlnet_xl_kolors_img2img.py create mode 100644 examples/community/pipeline_controlnet_xl_kolors_inpaint.py create mode 100644 examples/community/pipeline_kolors_inpainting.py diff --git a/examples/community/pipeline_controlnet_xl_kolors.py b/examples/community/pipeline_controlnet_xl_kolors.py new file mode 100644 index 000000000000..04b33260f97c --- /dev/null +++ b/examples/community/pipeline_controlnet_xl_kolors.py @@ -0,0 +1,1306 @@ +# Copyright 2024 The HuggingFace Team. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + + +import inspect +from typing import Any, Callable, Dict, List, Optional, Tuple, Union + +import numpy as np +import PIL.Image +import torch +import torch.nn.functional as F +from transformers import ( + CLIPImageProcessor, + CLIPVisionModelWithProjection, +) + +from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback +from diffusers.image_processor import PipelineImageInput, VaeImageProcessor +from diffusers.loaders import ( + FromSingleFileMixin, + IPAdapterMixin, + StableDiffusionXLLoraLoaderMixin, + TextualInversionLoaderMixin +) +from diffusers.models import AutoencoderKL, ImageProjection, UNet2DConditionModel +from diffusers.models.attention_processor import ( + AttnProcessor2_0, + XFormersAttnProcessor, +) +from diffusers.schedulers import KarrasDiffusionSchedulers +from diffusers.utils import ( + is_invisible_watermark_available, + replace_example_docstring, + deprecate, + logging, +) +from diffusers.utils.torch_utils import is_compiled_module, randn_tensor +from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin +from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput + +from diffusers.models import ControlNetModel, MultiControlNetModel + +from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer + +if is_invisible_watermark_available(): + from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker + +logger = logging.get_logger(__name__) # pylint: disable=invalid-name + + +EXAMPLE_DOC_STRING = """ + Examples: + ```py + >>> import torch + >>> from diffusers import KolorsControlNetPipeline, ControlNetModel + >>> from diffusers.utils import load_image + >>> import numpy as np + >>> import cv2 + >>> from PIL import Image + + >>> prompt = "aerial view, a futuristic research complex in a bright foggy jungle, hard lighting" + >>> negative_prompt = "low quality, bad quality, sketches" + + >>> # download an image + >>> image = load_image( + ... "https://hf.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/hf-logo.png" + ... ) + + >>> # initialize the models and pipeline + >>> controlnet_conditioning_scale = 0.5 # recommended for good generalization + >>> controlnet = ControlNetModel.from_pretrained( + ... "Kwai-Kolors/Kolors-ControlNet-Canny", torch_dtype=torch.float16 + ... ) + + >>> pipe = KolorsControlNetPipeline.from_pretrained( + ... "Kwai-Kolors/Kolors-diffusers", controlnet=controlnet, torch_dtype=torch.float16 + ... ) + >>> pipe.enable_model_cpu_offload() + + >>> # get canny image + >>> image = np.array(image) + >>> image = cv2.Canny(image, 100, 200) + >>> image = image[:, :, None] + >>> image = np.concatenate([image, image, image], axis=2) + >>> canny_image = Image.fromarray(image) + + >>> # generate image + >>> image = pipe( + ... prompt, controlnet_conditioning_scale=controlnet_conditioning_scale, image=canny_image + ... ).images[0] + ``` +""" + + +# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents +def retrieve_latents( + encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" +): + if hasattr(encoder_output, "latent_dist") and sample_mode == "sample": + return encoder_output.latent_dist.sample(generator) + elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax": + return encoder_output.latent_dist.mode() + elif hasattr(encoder_output, "latents"): + return encoder_output.latents + else: + raise AttributeError("Could not access latents of provided encoder_output") + + +class KolorsControlNetPipeline( + DiffusionPipeline, + StableDiffusionMixin, + StableDiffusionXLLoraLoaderMixin, + FromSingleFileMixin, + IPAdapterMixin, +): + r""" + Pipeline for image-to-image generation using Kolors with ControlNet guidance. + + This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the + library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.) + + The pipeline also inherits the following loading methods: + - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.safetensors` files + - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights + - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights + - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters + + Args: + vae ([`AutoencoderKL`]): + Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. + text_encoder ([`ChatGLMModel`]): + Frozen text-encoder. Kolors uses [ChatGLM3-6B](https://huggingface.co/THUDM/chatglm3-6b). + tokenizer (`ChatGLMTokenizer`): + Tokenizer of class + [ChatGLMTokenizer](https://huggingface.co/THUDM/chatglm3-6b/blob/main/tokenization_chatglm.py). + unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents. + controlnet ([`ControlNetModel`] or `List[ControlNetModel]`): + Provides additional conditioning to the unet during the denoising process. If you set multiple ControlNets + as a list, the outputs from each ControlNet are added together to create one combined additional + conditioning. + scheduler ([`SchedulerMixin`]): + A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of + [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`]. + requires_aesthetics_score (`bool`, *optional*, defaults to `"False"`): + Whether the `unet` requires an `aesthetic_score` condition to be passed during inference. + force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`): + Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of + `Kwai-Kolors/Kolors-diffusers`. + feature_extractor ([`~transformers.CLIPImageProcessor`]): + A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`. + """ + + model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae" + + _optional_components = [ + "tokenizer", + "text_encoder", + "feature_extractor", + "image_encoder", + ] + _callback_tensor_inputs = [ + "latents", + "prompt_embeds", + "negative_prompt_embeds", + "add_text_embeds", + "add_time_ids", + "negative_pooled_prompt_embeds", + "add_neg_time_ids", + ] + + def __init__( + self, + vae: AutoencoderKL, + text_encoder: ChatGLMModel, + tokenizer: ChatGLMTokenizer, + unet: UNet2DConditionModel, + controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel], + scheduler: KarrasDiffusionSchedulers, + requires_aesthetics_score: bool = False, + force_zeros_for_empty_prompt: bool = True, + feature_extractor: CLIPImageProcessor = None, + image_encoder: CLIPVisionModelWithProjection = None, + add_watermarker: Optional[bool] = None, + ): + super().__init__() + + if isinstance(controlnet, (list, tuple)): + controlnet = MultiControlNetModel(controlnet) + + self.register_modules( + vae=vae, + text_encoder=text_encoder, + tokenizer=tokenizer, + unet=unet, + controlnet=controlnet, + scheduler=scheduler, + feature_extractor=feature_extractor, + image_encoder=image_encoder, + ) + self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) + self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True) + self.control_image_processor = VaeImageProcessor( + vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False + ) + + if add_watermarker: + self.watermark = StableDiffusionXLWatermarker() + else: + self.watermark = None + + self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt) + self.register_to_config(requires_aesthetics_score=requires_aesthetics_score) + + def encode_prompt( + self, + prompt, + device: Optional[torch.device] = None, + num_images_per_prompt: int = 1, + do_classifier_free_guidance: bool = True, + negative_prompt=None, + prompt_embeds: Optional[torch.FloatTensor] = None, + negative_prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + lora_scale: Optional[float] = None, + ): + r""" + Encodes the prompt into text encoder hidden states. + + Args: + prompt (`str` or `List[str]`, *optional*): + prompt to be encoded + device: (`torch.device`): + torch device + num_images_per_prompt (`int`): + number of images that should be generated per prompt + do_classifier_free_guidance (`bool`): + whether to use classifier free guidance or not + negative_prompt (`str` or `List[str]`, *optional*): + The prompt or prompts not to guide the image generation. If not defined, one has to pass + `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is + less than `1`). + prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + negative_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input + argument. + pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` + input argument. + lora_scale (`float`, *optional*): + A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. + """ + device = device or self._execution_device + + # set lora scale so that monkey patched LoRA + # function of text encoder can correctly access it + if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin): + self._lora_scale = lora_scale + + if prompt is not None and isinstance(prompt, str): + batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + # Define tokenizers and text encoders + tokenizers = [self.tokenizer] + text_encoders = [self.text_encoder] + + if prompt_embeds is None: + # textual inversion: procecss multi-vector tokens if necessary + prompt_embeds_list = [] + for tokenizer, text_encoder in zip(tokenizers, text_encoders): + if isinstance(self, TextualInversionLoaderMixin): + prompt = self.maybe_convert_prompt(prompt, tokenizer) + + text_inputs = tokenizer( + prompt, + padding="max_length", + max_length=256, + truncation=True, + return_tensors="pt", + ).to(self._execution_device) + output = text_encoder( + input_ids=text_inputs['input_ids'], + attention_mask=text_inputs['attention_mask'], + position_ids=text_inputs['position_ids'], + output_hidden_states=True) + prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() + pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] + bs_embed, seq_len, _ = prompt_embeds.shape + prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) + prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1) + + prompt_embeds_list.append(prompt_embeds) + + prompt_embeds = prompt_embeds_list[0] + + # get unconditional embeddings for classifier free guidance + zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt + if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt: + negative_prompt_embeds = torch.zeros_like(prompt_embeds) + negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds) + elif do_classifier_free_guidance and negative_prompt_embeds is None: + uncond_tokens: List[str] + if negative_prompt is None: + uncond_tokens = [""] * batch_size + elif prompt is not None and type(prompt) is not type(negative_prompt): + raise TypeError( + f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" + f" {type(prompt)}." + ) + elif isinstance(negative_prompt, str): + uncond_tokens = [negative_prompt] + elif batch_size != len(negative_prompt): + raise ValueError( + f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" + f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" + " the batch size of `prompt`." + ) + else: + uncond_tokens = negative_prompt + + negative_prompt_embeds_list = [] + for tokenizer, text_encoder in zip(tokenizers, text_encoders): + # textual inversion: procecss multi-vector tokens if necessary + if isinstance(self, TextualInversionLoaderMixin): + uncond_tokens = self.maybe_convert_prompt(uncond_tokens, tokenizer) + + max_length = prompt_embeds.shape[1] + uncond_input = tokenizer( + uncond_tokens, + padding="max_length", + max_length=max_length, + truncation=True, + return_tensors="pt", + ).to(self._execution_device) + output = text_encoder( + input_ids=uncond_input['input_ids'], + attention_mask=uncond_input['attention_mask'], + position_ids=uncond_input['position_ids'], + output_hidden_states=True) + negative_prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() + negative_pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] + + if do_classifier_free_guidance: + # duplicate unconditional embeddings for each generation per prompt, using mps friendly method + seq_len = negative_prompt_embeds.shape[1] + + negative_prompt_embeds = negative_prompt_embeds.to(dtype=text_encoder.dtype, device=device) + + negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1) + negative_prompt_embeds = negative_prompt_embeds.view( + batch_size * num_images_per_prompt, seq_len, -1 + ) + + negative_prompt_embeds_list.append(negative_prompt_embeds) + + negative_prompt_embeds = negative_prompt_embeds_list[0] + + bs_embed = pooled_prompt_embeds.shape[0] + pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( + bs_embed * num_images_per_prompt, -1 + ) + if do_classifier_free_guidance: + negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( + bs_embed * num_images_per_prompt, -1 + ) + + return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds + + def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None): + dtype = next(self.image_encoder.parameters()).dtype + + if not isinstance(image, torch.Tensor): + image = self.feature_extractor(image, return_tensors="pt").pixel_values + + image = image.to(device=device, dtype=dtype) + if output_hidden_states: + image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2] + image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_enc_hidden_states = self.image_encoder( + torch.zeros_like(image), output_hidden_states=True + ).hidden_states[-2] + uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave( + num_images_per_prompt, dim=0 + ) + return image_enc_hidden_states, uncond_image_enc_hidden_states + else: + image_embeds = self.image_encoder(image).image_embeds + image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_embeds = torch.zeros_like(image_embeds) + + return image_embeds, uncond_image_embeds + + def prepare_ip_adapter_image_embeds( + self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance + ): + image_embeds = [] + if do_classifier_free_guidance: + negative_image_embeds = [] + if ip_adapter_image_embeds is None: + if not isinstance(ip_adapter_image, list): + ip_adapter_image = [ip_adapter_image] + + if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers): + raise ValueError( + f"`ip_adapter_image` must have same length as the number of IP Adapters. " + f"Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters." + ) + + for single_ip_adapter_image, image_proj_layer in zip( + ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers + ): + output_hidden_state = not isinstance(image_proj_layer, ImageProjection) + single_image_embeds, single_negative_image_embeds = self.encode_image( + single_ip_adapter_image, device, 1, output_hidden_state + ) + + image_embeds.append(single_image_embeds[None, :]) + if do_classifier_free_guidance: + negative_image_embeds.append(single_negative_image_embeds[None, :]) + else: + for single_image_embeds in ip_adapter_image_embeds: + if do_classifier_free_guidance: + single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2) + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image + def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None): + dtype = next(self.image_encoder.parameters()).dtype + + if not isinstance(image, torch.Tensor): + image = self.feature_extractor(image, return_tensors="pt").pixel_values + + image = image.to(device=device, dtype=dtype) + if output_hidden_states: + image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2] + image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_enc_hidden_states = self.image_encoder( + torch.zeros_like(image), output_hidden_states=True + ).hidden_states[-2] + uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave( + num_images_per_prompt, dim=0 + ) + return image_enc_hidden_states, uncond_image_enc_hidden_states + else: + image_embeds = self.image_encoder(image).image_embeds + image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_embeds = torch.zeros_like(image_embeds) + + return image_embeds, uncond_image_embeds + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs + def prepare_extra_step_kwargs(self, generator, eta): + # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature + # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. + # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 + # and should be between [0, 1] + + accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) + extra_step_kwargs = {} + if accepts_eta: + extra_step_kwargs["eta"] = eta + + # check if the scheduler accepts generator + accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) + if accepts_generator: + extra_step_kwargs["generator"] = generator + return extra_step_kwargs + + def check_inputs( + self, + prompt, + image, + num_inference_steps, + callback_steps, + negative_prompt=None, + prompt_embeds=None, + negative_prompt_embeds=None, + pooled_prompt_embeds=None, + negative_pooled_prompt_embeds=None, + ip_adapter_image=None, + ip_adapter_image_embeds=None, + controlnet_conditioning_scale=1.0, + control_guidance_start=0.0, + control_guidance_end=1.0, + callback_on_step_end_tensor_inputs=None, + ): + if num_inference_steps is None: + raise ValueError("`num_inference_steps` cannot be None.") + elif not isinstance(num_inference_steps, int) or num_inference_steps <= 0: + raise ValueError( + f"`num_inference_steps` has to be a positive integer but is {num_inference_steps} of type" + f" {type(num_inference_steps)}." + ) + + if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0): + raise ValueError( + f"`callback_steps` has to be a positive integer but is {callback_steps} of type" + f" {type(callback_steps)}." + ) + + if callback_on_step_end_tensor_inputs is not None and not all( + k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs + ): + raise ValueError( + f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" + ) + + if prompt is not None and prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to" + " only forward one of the two." + ) + elif prompt is None and prompt_embeds is None: + raise ValueError( + "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." + ) + elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): + raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") + + if negative_prompt is not None and negative_prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:" + f" {negative_prompt_embeds}. Please make sure to only forward one of the two." + ) + + if prompt_embeds is not None and negative_prompt_embeds is not None: + if prompt_embeds.shape != negative_prompt_embeds.shape: + raise ValueError( + "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but" + f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`" + f" {negative_prompt_embeds.shape}." + ) + + if prompt_embeds is not None and pooled_prompt_embeds is None: + raise ValueError( + "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`." + ) + + if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None: + raise ValueError( + "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`." + ) + + # `prompt` needs more sophisticated handling when there are multiple + # conditionings. + if isinstance(self.controlnet, MultiControlNetModel): + if isinstance(prompt, list): + logger.warning( + f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}" + " prompts. The conditionings will be fixed across the prompts." + ) + + is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance( + self.controlnet, torch._dynamo.eval_frame.OptimizedModule + ) + + # Check `controlnet_conditioning_scale` + if ( + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) + ): + if not isinstance(controlnet_conditioning_scale, float): + raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.") + elif ( + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + ): + if isinstance(controlnet_conditioning_scale, list): + if any(isinstance(i, list) for i in controlnet_conditioning_scale): + raise ValueError("A single batch of multiple conditionings are supported at the moment.") + elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len( + self.controlnet.nets + ): + raise ValueError( + "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have" + " the same length as the number of controlnets" + ) + else: + assert False + + if not isinstance(control_guidance_start, (tuple, list)): + control_guidance_start = [control_guidance_start] + + if not isinstance(control_guidance_end, (tuple, list)): + control_guidance_end = [control_guidance_end] + + if len(control_guidance_start) != len(control_guidance_end): + raise ValueError( + f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list." + ) + + if isinstance(self.controlnet, MultiControlNetModel): + if len(control_guidance_start) != len(self.controlnet.nets): + raise ValueError( + f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}." + ) + + for start, end in zip(control_guidance_start, control_guidance_end): + if start >= end: + raise ValueError( + f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}." + ) + if start < 0.0: + raise ValueError(f"control guidance start: {start} can't be smaller than 0.") + if end > 1.0: + raise ValueError(f"control guidance end: {end} can't be larger than 1.0.") + + if ip_adapter_image is not None and ip_adapter_image_embeds is not None: + raise ValueError( + "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined." + ) + + if ip_adapter_image_embeds is not None: + if not isinstance(ip_adapter_image_embeds, list): + raise ValueError( + f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}" + ) + elif ip_adapter_image_embeds[0].ndim not in [3, 4]: + raise ValueError( + f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D" + ) + + # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.check_image + def check_image(self, image, prompt, prompt_embeds): + image_is_pil = isinstance(image, PIL.Image.Image) + image_is_tensor = isinstance(image, torch.Tensor) + image_is_np = isinstance(image, np.ndarray) + image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image) + image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor) + image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray) + + if ( + not image_is_pil + and not image_is_tensor + and not image_is_np + and not image_is_pil_list + and not image_is_tensor_list + and not image_is_np_list + ): + raise TypeError( + f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}" + ) + + if image_is_pil: + image_batch_size = 1 + else: + image_batch_size = len(image) + + if prompt is not None and isinstance(prompt, str): + prompt_batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + prompt_batch_size = len(prompt) + elif prompt_embeds is not None: + prompt_batch_size = prompt_embeds.shape[0] + + if image_batch_size != 1 and image_batch_size != prompt_batch_size: + raise ValueError( + f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}" + ) + + # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.prepare_image + def prepare_control_image( + self, + image, + width, + height, + batch_size, + num_images_per_prompt, + device, + dtype, + do_classifier_free_guidance=False, + guess_mode=False, + ): + image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) + image_batch_size = image.shape[0] + + if image_batch_size == 1: + repeat_by = batch_size + else: + # image batch size is the same as prompt batch size + repeat_by = num_images_per_prompt + + image = image.repeat_interleave(repeat_by, dim=0) + + image = image.to(device=device, dtype=dtype) + + if do_classifier_free_guidance and not guess_mode: + image = torch.cat([image] * 2) + + return image + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents + def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None): + shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) + if isinstance(generator, list) and len(generator) != batch_size: + raise ValueError( + f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" + f" size of {batch_size}. Make sure the batch size matches the length of the generators." + ) + + if latents is None: + latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype) + else: + latents = latents.to(device) + + # scale the initial noise by the standard deviation required by the scheduler + latents = latents * self.scheduler.init_noise_sigma + return latents + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents + def prepare_latents_t2i(self, batch_size, num_channels_latents, height, width, dtype, device, generator, + latents=None): + shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) + if isinstance(generator, list) and len(generator) != batch_size: + raise ValueError( + f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" + f" size of {batch_size}. Make sure the batch size matches the length of the generators." + ) + + if latents is None: + latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype) + else: + latents = latents.to(device) + + # scale the initial noise by the standard deviation required by the scheduler + latents = latents * self.scheduler.init_noise_sigma + return latents + + def _get_add_time_ids(self, original_size, crops_coords_top_left, target_size, dtype): + add_time_ids = list(original_size + crops_coords_top_left + target_size) + + passed_add_embed_dim = ( + self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 + ) + expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features + + if expected_add_embed_dim != passed_add_embed_dim: + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`." + ) + + add_time_ids = torch.tensor([add_time_ids], dtype=dtype) + return add_time_ids + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae + def upcast_vae(self): + dtype = self.vae.dtype + self.vae.to(dtype=torch.float32) + use_torch_2_0_or_xformers = isinstance( + self.vae.decoder.mid_block.attentions[0].processor, + ( + AttnProcessor2_0, + XFormersAttnProcessor, + ), + ) + # if xformers or torch_2_0 is used attention block does not need + # to be in float32 which can save lots of memory + if use_torch_2_0_or_xformers: + self.vae.post_quant_conv.to(dtype) + self.vae.decoder.conv_in.to(dtype) + self.vae.decoder.mid_block.to(dtype) + + @property + def guidance_scale(self): + return self._guidance_scale + + # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2) + # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1` + # corresponds to doing no classifier free guidance. + @property + def do_classifier_free_guidance(self): + return self._guidance_scale > 1 + + @property + def cross_attention_kwargs(self): + return self._cross_attention_kwargs + + @property + def num_timesteps(self): + return self._num_timesteps + + @torch.no_grad() + @replace_example_docstring(EXAMPLE_DOC_STRING) + def __call__( + self, + prompt: Union[str, List[str]] = None, + control_image: PipelineImageInput = None, + height: Optional[int] = None, + width: Optional[int] = None, + num_inference_steps: int = 50, + guidance_scale: float = 5.0, + negative_prompt: Optional[Union[str, List[str]]] = None, + num_images_per_prompt: Optional[int] = 1, + eta: float = 0.0, + guess_mode: bool = False, + generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, + latents: Optional[torch.Tensor] = None, + prompt_embeds: Optional[torch.Tensor] = None, + negative_prompt_embeds: Optional[torch.Tensor] = None, + pooled_prompt_embeds: Optional[torch.Tensor] = None, + negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, + ip_adapter_image: Optional[PipelineImageInput] = None, + ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, + output_type: Optional[str] = "pil", + return_dict: bool = True, + cross_attention_kwargs: Optional[Dict[str, Any]] = None, + controlnet_conditioning_scale: Union[float, List[float]] = 0.8, + control_guidance_start: Union[float, List[float]] = 0.0, + control_guidance_end: Union[float, List[float]] = 1.0, + original_size: Tuple[int, int] = None, + crops_coords_top_left: Tuple[int, int] = (0, 0), + target_size: Tuple[int, int] = None, + callback_on_step_end: Optional[ + Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] + ] = None, + callback_on_step_end_tensor_inputs: List[str] = ["latents"], + **kwargs, + ): + r""" + Function invoked when calling the pipeline for generation. + + Args: + prompt (`str` or `List[str]`, *optional*): + The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`. + instead. + control_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: + `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): + The ControlNet input condition. ControlNet uses this input condition to generate guidance to Unet. If + the type is specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also + be accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height + and/or width are passed, `image` is resized according to them. If multiple ControlNets are specified in + init, images must be passed as a list such that each element of the list can be correctly batched for + input to a single controlnet. + height (`int`, *optional*, defaults to the size of control_image): + The height in pixels of the generated image. Anything below 512 pixels won't work well for + [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) + and checkpoints that are not specifically fine-tuned on low resolutions. + width (`int`, *optional*, defaults to the size of control_image): + The width in pixels of the generated image. Anything below 512 pixels won't work well for + [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) + and checkpoints that are not specifically fine-tuned on low resolutions. + strength (`float`, *optional*, defaults to 0.8): + Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a + starting point and more noise is added the higher the `strength`. The number of denoising steps depends + on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising + process runs for the full number of iterations specified in `num_inference_steps`. A value of 1 + essentially ignores `image`. + num_inference_steps (`int`, *optional*, defaults to 50): + The number of denoising steps. More denoising steps usually lead to a higher quality image at the + expense of slower inference. + guidance_scale (`float`, *optional*, defaults to 7.5): + Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598). + `guidance_scale` is defined as `w` of equation 2. of [Imagen + Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale > + 1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, + usually at the expense of lower image quality. + negative_prompt (`str` or `List[str]`, *optional*): + The prompt or prompts not to guide the image generation. If not defined, one has to pass + `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is + less than `1`). + num_images_per_prompt (`int`, *optional*, defaults to 1): + The number of images to generate per prompt. + eta (`float`, *optional*, defaults to 0.0): + Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to + [`schedulers.DDIMScheduler`], will be ignored for others. + generator (`torch.Generator` or `List[torch.Generator]`, *optional*): + One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html) + to make generation deterministic. + latents (`torch.Tensor`, *optional*): + Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image + generation. Can be used to tweak the same generation with different prompts. If not provided, a latents + tensor will ge generated by sampling using the supplied random `generator`. + prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + negative_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input + argument. + pooled_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + negative_pooled_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` + input argument. + ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters. + ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*): + Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of + IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should + contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not + provided, embeddings are computed from the `ip_adapter_image` input argument. + output_type (`str`, *optional*, defaults to `"pil"`): + The output format of the generate image. Choose between + [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`. + return_dict (`bool`, *optional*, defaults to `True`): + Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a + plain tuple. + cross_attention_kwargs (`dict`, *optional*): + A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under + `self.processor` in + [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). + controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0): + The outputs of the controlnet are multiplied by `controlnet_conditioning_scale` before they are added + to the residual in the original unet. If multiple ControlNets are specified in init, you can set the + corresponding scale as a list. + control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0): + The percentage of total steps at which the controlnet starts applying. + control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0): + The percentage of total steps at which the controlnet stops applying. + original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled. + `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as + explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): + `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position + `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting + `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + For most cases, `target_size` should be set to the desired height and width of the generated image. If + not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in + section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*): + A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of + each denoising step during the inference. with the following arguments: `callback_on_step_end(self: + DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a + list of all tensors as specified by `callback_on_step_end_tensor_inputs`. + callback_on_step_end_tensor_inputs (`List`, *optional*): + The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list + will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the + `._callback_tensor_inputs` attribute of your pipeline class. + + Examples: + + Returns: + [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`: + [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple` + containing the output images. + """ + + callback = kwargs.pop("callback", None) + callback_steps = kwargs.pop("callback_steps", None) + + if callback is not None: + deprecate( + "callback", + "1.0.0", + "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", + ) + if callback_steps is not None: + deprecate( + "callback_steps", + "1.0.0", + "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", + ) + + if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)): + callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs + + controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet + + # align format for control guidance + if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list): + control_guidance_start = len(control_guidance_end) * [control_guidance_start] + elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list): + control_guidance_end = len(control_guidance_start) * [control_guidance_end] + elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list): + mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1 + control_guidance_start, control_guidance_end = ( + mult * [control_guidance_start], + mult * [control_guidance_end], + ) + + # from IPython import embed; embed() + # 1. Check inputs. Raise error if not correct + self.check_inputs( + prompt, + control_image, + num_inference_steps, + callback_steps, + negative_prompt, + prompt_embeds, + negative_prompt_embeds, + pooled_prompt_embeds, + negative_pooled_prompt_embeds, + ip_adapter_image, + ip_adapter_image_embeds, + controlnet_conditioning_scale, + control_guidance_start, + control_guidance_end, + callback_on_step_end_tensor_inputs, + ) + + self._guidance_scale = guidance_scale + self._cross_attention_kwargs = cross_attention_kwargs + + # 2. Define call parameters + if prompt is not None and isinstance(prompt, str): + batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + device = self._execution_device + + if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float): + controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets) + + # 3.1. Encode input prompt + text_encoder_lora_scale = ( + self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None + ) + ( + prompt_embeds, + negative_prompt_embeds, + pooled_prompt_embeds, + negative_pooled_prompt_embeds, + ) = self.encode_prompt( + prompt, + device, + num_images_per_prompt, + self.do_classifier_free_guidance, + negative_prompt, + prompt_embeds=prompt_embeds, + negative_prompt_embeds=negative_prompt_embeds, + pooled_prompt_embeds=pooled_prompt_embeds, + negative_pooled_prompt_embeds=negative_pooled_prompt_embeds, + lora_scale=text_encoder_lora_scale, + ) + + # 3.2 Encode ip_adapter_image + if ip_adapter_image is not None or ip_adapter_image_embeds is not None: + image_embeds = self.prepare_ip_adapter_image_embeds( + ip_adapter_image, + ip_adapter_image_embeds, + device, + batch_size * num_images_per_prompt, + self.do_classifier_free_guidance, + ) + + if isinstance(controlnet, ControlNetModel): + control_image = self.prepare_control_image( + image=control_image, + width=width, + height=height, + batch_size=batch_size * num_images_per_prompt, + num_images_per_prompt=num_images_per_prompt, + device=device, + dtype=controlnet.dtype, + do_classifier_free_guidance=self.do_classifier_free_guidance, + guess_mode=guess_mode, + ) + height, width = control_image.shape[-2:] + elif isinstance(controlnet, MultiControlNetModel): + control_images = [] + + for control_image_ in control_image: + control_image_ = self.prepare_control_image( + image=control_image_, + width=width, + height=height, + batch_size=batch_size * num_images_per_prompt, + num_images_per_prompt=num_images_per_prompt, + device=device, + dtype=controlnet.dtype, + do_classifier_free_guidance=self.do_classifier_free_guidance, + guess_mode=guess_mode, + ) + + control_images.append(control_image_) + + control_image = control_images + height, width = control_image[0].shape[-2:] + else: + assert False + + # 4. Prepare timesteps + self.scheduler.set_timesteps(num_inference_steps, device=device) + + timesteps = self.scheduler.timesteps + + # 5. Prepare latent variables + num_channels_latents = self.unet.config.in_channels + latents = self.prepare_latents( + batch_size * num_images_per_prompt, + num_channels_latents, + height, + width, + prompt_embeds.dtype, + device, + generator, + latents, + ) + + # 6.5 Optionally get Guidance Scale Embedding + timestep_cond = None + if self.unet.config.time_cond_proj_dim is not None: + guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt) + timestep_cond = self.get_guidance_scale_embedding( + guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim + ).to(device=device, dtype=latents.dtype) + + # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline + extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta) + + # 7.1 Create tensor stating which controlnets to keep + controlnet_keep = [] + for i in range(len(timesteps)): + keeps = [ + 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e) + for s, e in zip(control_guidance_start, control_guidance_end) + ] + controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps) + + # 7.2 Prepare added time ids & embeddings + if isinstance(control_image, list): + original_size = original_size or control_image[0].shape[-2:] + else: + original_size = original_size or control_image.shape[-2:] + target_size = target_size or (height, width) + + # 7. Prepare added time ids & embeddings + add_text_embeds = pooled_prompt_embeds + add_time_ids = self._get_add_time_ids( + original_size, crops_coords_top_left, target_size, dtype=prompt_embeds.dtype + ) + + if self.do_classifier_free_guidance: + prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) + add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0) + add_time_ids = torch.cat([add_time_ids, add_time_ids], dim=0) + + prompt_embeds = prompt_embeds.to(device) + add_text_embeds = add_text_embeds.to(device) + add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1) + + # patch diffusers controlnet instance forward, undo + # after denoising loop + + patched_cn_models = [] + if isinstance(self.controlnet, MultiControlNetModel): + cn_models_to_patch = self.controlnet.nets + else: + cn_models_to_patch = [self.controlnet] + + for cn_model in cn_models_to_patch: + cn_og_forward = cn_model.forward + + def _cn_patch_forward(*args, **kwargs): + encoder_hidden_states = kwargs['encoder_hidden_states'] + if cn_model.encoder_hid_proj is not None and cn_model.config.encoder_hid_dim_type == "text_proj": + encoder_hidden_states = cn_model.encoder_hid_proj(kwargs['encoder_hidden_states']) + kwargs.pop('encoder_hidden_states') + return cn_og_forward(*args, encoder_hidden_states=encoder_hidden_states, **kwargs) + + cn_model.forward = _cn_patch_forward + patched_cn_models.append((cn_model, cn_og_forward)) + + # 8. Denoising loop + num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order + try: + with self.progress_bar(total=num_inference_steps) as progress_bar: + for i, t in enumerate(timesteps): + # expand the latents if we are doing classifier free guidance + latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents + latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) + + added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids} + + # controlnet(s) inference + if guess_mode and self.do_classifier_free_guidance: + # Infer ControlNet only for the conditional batch. + control_model_input = latents + control_model_input = self.scheduler.scale_model_input(control_model_input, t) + controlnet_prompt_embeds = prompt_embeds.chunk(2)[1] + controlnet_added_cond_kwargs = { + "text_embeds": add_text_embeds.chunk(2)[1], + "time_ids": add_time_ids.chunk(2)[1], + } + else: + control_model_input = latent_model_input + controlnet_prompt_embeds = prompt_embeds + controlnet_added_cond_kwargs = added_cond_kwargs + + if isinstance(controlnet_keep[i], list): + cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])] + else: + controlnet_cond_scale = controlnet_conditioning_scale + if isinstance(controlnet_cond_scale, list): + controlnet_cond_scale = controlnet_cond_scale[0] + cond_scale = controlnet_cond_scale * controlnet_keep[i] + + down_block_res_samples, mid_block_res_sample = self.controlnet( + control_model_input, + t, + encoder_hidden_states=controlnet_prompt_embeds, + controlnet_cond=control_image, + conditioning_scale=cond_scale, + guess_mode=guess_mode, + added_cond_kwargs=controlnet_added_cond_kwargs, + return_dict=False, + ) + + if guess_mode and self.do_classifier_free_guidance: + # Infered ControlNet only for the conditional batch. + # To apply the output of ControlNet to both the unconditional and conditional batches, + # add 0 to the unconditional batch to keep it unchanged. + down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples] + mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample]) + + if ip_adapter_image is not None or ip_adapter_image_embeds is not None: + added_cond_kwargs["image_embeds"] = image_embeds + + # predict the noise residual + noise_pred = self.unet( + latent_model_input, + t, + encoder_hidden_states=prompt_embeds, + timestep_cond=timestep_cond, + cross_attention_kwargs=self.cross_attention_kwargs, + down_block_additional_residuals=down_block_res_samples, + mid_block_additional_residual=mid_block_res_sample, + added_cond_kwargs=added_cond_kwargs, + return_dict=False, + )[0] + + # perform guidance + if self.do_classifier_free_guidance: + noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) + noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) + + # compute the previous noisy sample x_t -> x_t-1 + latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] + + # call the callback, if provided + if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): + progress_bar.update() + if callback is not None and i % callback_steps == 0: + step_idx = i // getattr(self.scheduler, "order", 1) + callback(step_idx, t, latents) + + finally: + for cn_and_og in patched_cn_models: + cn_and_og[0].forward = cn_and_og[1] + + # If we do sequential model offloading, let's offload unet and controlnet + # manually for max memory savings + if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None: + self.unet.to("cpu") + self.controlnet.to("cpu") + torch.cuda.empty_cache() + torch.cuda.ipc_collect() + + if not output_type == "latent": + # make sure the VAE is in float32 mode, as it overflows in float16 + needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast + + if needs_upcasting: + self.upcast_vae() + latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype) + + latents = latents / self.vae.config.scaling_factor + image = self.vae.decode(latents, return_dict=False)[0] + + # cast back to fp16 if needed + if needs_upcasting: + self.vae.to(dtype=torch.float16) + else: + image = latents + return StableDiffusionXLPipelineOutput(images=image) + + image = self.image_processor.postprocess(image, output_type=output_type) + + # Offload all models + self.maybe_free_model_hooks() + + if not return_dict: + return (image,) + + return StableDiffusionXLPipelineOutput(images=image) diff --git a/examples/community/pipeline_controlnet_xl_kolors_img2img.py b/examples/community/pipeline_controlnet_xl_kolors_img2img.py new file mode 100644 index 000000000000..29be6d08a428 --- /dev/null +++ b/examples/community/pipeline_controlnet_xl_kolors_img2img.py @@ -0,0 +1,1443 @@ +# Copyright 2024 The HuggingFace Team. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + + +import inspect +from typing import Any, Callable, Dict, List, Optional, Tuple, Union + +import PIL.Image +import numpy as np +import torch +import torch.nn.functional as F +from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback +from diffusers.image_processor import PipelineImageInput, VaeImageProcessor +from diffusers.loaders import ( + FromSingleFileMixin, + IPAdapterMixin, + StableDiffusionXLLoraLoaderMixin, + TextualInversionLoaderMixin +) +from diffusers.models import AutoencoderKL, ImageProjection, UNet2DConditionModel +from diffusers.models import ControlNetModel, MultiControlNetModel +from diffusers.models.attention_processor import ( + AttnProcessor2_0, + XFormersAttnProcessor, +) + +from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin +from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput +from diffusers.schedulers import KarrasDiffusionSchedulers +from diffusers.utils import ( + is_invisible_watermark_available, + replace_example_docstring, + deprecate, + logging, +) +from diffusers.utils.torch_utils import is_compiled_module, randn_tensor +from transformers import ( + CLIPImageProcessor, + CLIPVisionModelWithProjection, +) + +from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer + +if is_invisible_watermark_available(): + from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker + +logger = logging.get_logger(__name__) # pylint: disable=invalid-name + + +EXAMPLE_DOC_STRING = """ + Examples: + ```py + >>> import torch + >>> import numpy as np + >>> from PIL import Image + + >>> from transformers import DPTImageProcessor, DPTForDepthEstimation + >>> from diffusers import ControlNetModel, KolorsControlNetImg2ImgPipeline + >>> from diffusers.utils import load_image + + >>> depth_estimator = DPTForDepthEstimation.from_pretrained("Intel/dpt-hybrid-midas").to("cuda") + >>> feature_extractor = DPTImageProcessor.from_pretrained("Intel/dpt-hybrid-midas") + >>> controlnet = ControlNetModel.from_pretrained( + ... "Kwai-Kolors/Kolors-ControlNet-Depth", + ... variant="fp16", + ... use_safetensors=True, + ... torch_dtype=torch.float16, + ... ) + >>> pipe = KolorsControlNetImg2ImgPipeline.from_pretrained( + ... "Kwai-Kolors/Kolors-diffusers", + ... controlnet=controlnet, + ... variant="fp16", + ... use_safetensors=True, + ... torch_dtype=torch.float16, + ... ) + >>> pipe.enable_model_cpu_offload() + + + >>> def get_depth_map(image): + ... image = feature_extractor(images=image, return_tensors="pt").pixel_values.to("cuda") + ... + ... with torch.no_grad(), torch.autocast("cuda"): + ... depth_map = depth_estimator(image).predicted_depth + ... + ... depth_map = torch.nn.functional.interpolate( + ... depth_map.unsqueeze(1), + ... size=(1024, 1024), + ... mode="bicubic", + ... align_corners=False, + ... ) + ... depth_min = torch.amin(depth_map, dim=[1, 2, 3], keepdim=True) + ... depth_max = torch.amax(depth_map, dim=[1, 2, 3], keepdim=True) + ... depth_map = (depth_map - depth_min) / (depth_max - depth_min) + ... image = torch.cat([depth_map] * 3, dim=1) + ... image = image.permute(0, 2, 3, 1).cpu().numpy()[0] + ... image = Image.fromarray((image * 255.0).clip(0, 255).astype(np.uint8)) + ... return image + + + >>> prompt = "A robot, 4k photo" + >>> image = load_image( + ... "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main" + ... "/kandinsky/cat.png" + ... ).resize((1024, 1024)) + >>> controlnet_conditioning_scale = 0.5 # recommended for good generalization + >>> depth_image = get_depth_map(image) + + >>> images = pipe( + ... prompt, + ... image=image, + ... control_image=depth_image, + ... strength=0.99, + ... num_inference_steps=50, + ... controlnet_conditioning_scale=controlnet_conditioning_scale, + ... ).images + >>> images[0].save(f"robot_cat.png") + ``` +""" + + +# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents +def retrieve_latents( + encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" +): + if hasattr(encoder_output, "latent_dist") and sample_mode == "sample": + return encoder_output.latent_dist.sample(generator) + elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax": + return encoder_output.latent_dist.mode() + elif hasattr(encoder_output, "latents"): + return encoder_output.latents + else: + raise AttributeError("Could not access latents of provided encoder_output") + + +class KolorsControlNetImg2ImgPipeline( + DiffusionPipeline, + StableDiffusionMixin, + StableDiffusionXLLoraLoaderMixin, + FromSingleFileMixin, + IPAdapterMixin, +): + r""" + Pipeline for image-to-image generation using Kolors with ControlNet guidance. + + This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the + library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.) + + The pipeline also inherits the following loading methods: + - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.safetensors` files + - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights + - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights + - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters + + Args: + vae ([`AutoencoderKL`]): + Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. + text_encoder ([`ChatGLMModel`]): + Frozen text-encoder. Kolors uses [ChatGLM3-6B](https://huggingface.co/THUDM/chatglm3-6b). + tokenizer (`ChatGLMTokenizer`): + Tokenizer of class + [ChatGLMTokenizer](https://huggingface.co/THUDM/chatglm3-6b/blob/main/tokenization_chatglm.py). + unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents. + controlnet ([`ControlNetModel`] or `List[ControlNetModel]`): + Provides additional conditioning to the unet during the denoising process. If you set multiple ControlNets + as a list, the outputs from each ControlNet are added together to create one combined additional + conditioning. + scheduler ([`SchedulerMixin`]): + A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of + [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`]. + requires_aesthetics_score (`bool`, *optional*, defaults to `"False"`): + Whether the `unet` requires an `aesthetic_score` condition to be passed during inference. Also see the + config of `stabilityai/stable-diffusion-xl-refiner-1-0`. + force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`): + Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of + `Kwai-Kolors/Kolors-diffusers`. + feature_extractor ([`~transformers.CLIPImageProcessor`]): + A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`. + """ + + model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae" + + _optional_components = [ + "tokenizer", + "text_encoder", + "feature_extractor", + "image_encoder", + ] + _callback_tensor_inputs = [ + "latents", + "prompt_embeds", + "negative_prompt_embeds", + "add_text_embeds", + "add_time_ids", + "negative_pooled_prompt_embeds", + "add_neg_time_ids", + ] + + def __init__( + self, + vae: AutoencoderKL, + text_encoder: ChatGLMModel, + tokenizer: ChatGLMTokenizer, + unet: UNet2DConditionModel, + controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel], + scheduler: KarrasDiffusionSchedulers, + requires_aesthetics_score: bool = False, + force_zeros_for_empty_prompt: bool = True, + feature_extractor: CLIPImageProcessor = None, + image_encoder: CLIPVisionModelWithProjection = None, + add_watermarker: Optional[bool] = None, + ): + super().__init__() + + if isinstance(controlnet, (list, tuple)): + controlnet = MultiControlNetModel(controlnet) + + self.register_modules( + vae=vae, + text_encoder=text_encoder, + tokenizer=tokenizer, + unet=unet, + controlnet=controlnet, + scheduler=scheduler, + feature_extractor=feature_extractor, + image_encoder=image_encoder, + ) + self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) + self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True) + self.control_image_processor = VaeImageProcessor( + vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False + ) + + if add_watermarker: + self.watermark = StableDiffusionXLWatermarker() + else: + self.watermark = None + + self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt) + self.register_to_config(requires_aesthetics_score=requires_aesthetics_score) + + def encode_prompt( + self, + prompt, + device: Optional[torch.device] = None, + num_images_per_prompt: int = 1, + do_classifier_free_guidance: bool = True, + negative_prompt=None, + prompt_embeds: Optional[torch.FloatTensor] = None, + negative_prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + lora_scale: Optional[float] = None, + ): + r""" + Encodes the prompt into text encoder hidden states. + + Args: + prompt (`str` or `List[str]`, *optional*): + prompt to be encoded + device: (`torch.device`): + torch device + num_images_per_prompt (`int`): + number of images that should be generated per prompt + do_classifier_free_guidance (`bool`): + whether to use classifier free guidance or not + negative_prompt (`str` or `List[str]`, *optional*): + The prompt or prompts not to guide the image generation. If not defined, one has to pass + `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is + less than `1`). + prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + negative_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input + argument. + pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` + input argument. + lora_scale (`float`, *optional*): + A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. + """ + # from IPython import embed; embed(); exit() + device = device or self._execution_device + + # set lora scale so that monkey patched LoRA + # function of text encoder can correctly access it + if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin): + self._lora_scale = lora_scale + + if prompt is not None and isinstance(prompt, str): + batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + # Define tokenizers and text encoders + tokenizers = [self.tokenizer] + text_encoders = [self.text_encoder] + + if prompt_embeds is None: + # textual inversion: procecss multi-vector tokens if necessary + prompt_embeds_list = [] + for tokenizer, text_encoder in zip(tokenizers, text_encoders): + if isinstance(self, TextualInversionLoaderMixin): + prompt = self.maybe_convert_prompt(prompt, tokenizer) + + text_inputs = tokenizer( + prompt, + padding="max_length", + max_length=256, + truncation=True, + return_tensors="pt", + ).to(self._execution_device) + output = text_encoder( + input_ids=text_inputs['input_ids'], + attention_mask=text_inputs['attention_mask'], + position_ids=text_inputs['position_ids'], + output_hidden_states=True) + prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() + pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] + bs_embed, seq_len, _ = prompt_embeds.shape + prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) + prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1) + + prompt_embeds_list.append(prompt_embeds) + + # prompt_embeds = torch.concat(prompt_embeds_list, dim=-1) + prompt_embeds = prompt_embeds_list[0] + + # get unconditional embeddings for classifier free guidance + zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt + if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt: + negative_prompt_embeds = torch.zeros_like(prompt_embeds) + negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds) + elif do_classifier_free_guidance and negative_prompt_embeds is None: + # negative_prompt = negative_prompt or "" + uncond_tokens: List[str] + if negative_prompt is None: + uncond_tokens = [""] * batch_size + elif prompt is not None and type(prompt) is not type(negative_prompt): + raise TypeError( + f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" + f" {type(prompt)}." + ) + elif isinstance(negative_prompt, str): + uncond_tokens = [negative_prompt] + elif batch_size != len(negative_prompt): + raise ValueError( + f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" + f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" + " the batch size of `prompt`." + ) + else: + uncond_tokens = negative_prompt + + negative_prompt_embeds_list = [] + for tokenizer, text_encoder in zip(tokenizers, text_encoders): + # textual inversion: procecss multi-vector tokens if necessary + if isinstance(self, TextualInversionLoaderMixin): + uncond_tokens = self.maybe_convert_prompt(uncond_tokens, tokenizer) + + max_length = prompt_embeds.shape[1] + uncond_input = tokenizer( + uncond_tokens, + padding="max_length", + max_length=max_length, + truncation=True, + return_tensors="pt", + ).to(self._execution_device) + output = text_encoder( + input_ids=uncond_input['input_ids'], + attention_mask=uncond_input['attention_mask'], + position_ids=uncond_input['position_ids'], + output_hidden_states=True) + negative_prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() + negative_pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] + + if do_classifier_free_guidance: + # duplicate unconditional embeddings for each generation per prompt, using mps friendly method + seq_len = negative_prompt_embeds.shape[1] + + negative_prompt_embeds = negative_prompt_embeds.to(dtype=text_encoder.dtype, device=device) + + negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1) + negative_prompt_embeds = negative_prompt_embeds.view( + batch_size * num_images_per_prompt, seq_len, -1 + ) + + # For classifier free guidance, we need to do two forward passes. + # Here we concatenate the unconditional and text embeddings into a single batch + # to avoid doing two forward passes + + negative_prompt_embeds_list.append(negative_prompt_embeds) + + # negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1) + negative_prompt_embeds = negative_prompt_embeds_list[0] + + bs_embed = pooled_prompt_embeds.shape[0] + pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( + bs_embed * num_images_per_prompt, -1 + ) + if do_classifier_free_guidance: + negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( + bs_embed * num_images_per_prompt, -1 + ) + + return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image + def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None): + dtype = next(self.image_encoder.parameters()).dtype + + if not isinstance(image, torch.Tensor): + image = self.feature_extractor(image, return_tensors="pt").pixel_values + + image = image.to(device=device, dtype=dtype) + if output_hidden_states: + image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2] + image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_enc_hidden_states = self.image_encoder( + torch.zeros_like(image), output_hidden_states=True + ).hidden_states[-2] + uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave( + num_images_per_prompt, dim=0 + ) + return image_enc_hidden_states, uncond_image_enc_hidden_states + else: + image_embeds = self.image_encoder(image).image_embeds + image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_embeds = torch.zeros_like(image_embeds) + + return image_embeds, uncond_image_embeds + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds + def prepare_ip_adapter_image_embeds( + self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance + ): + image_embeds = [] + if do_classifier_free_guidance: + negative_image_embeds = [] + if ip_adapter_image_embeds is None: + if not isinstance(ip_adapter_image, list): + ip_adapter_image = [ip_adapter_image] + + if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers): + raise ValueError( + f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters." + ) + + for single_ip_adapter_image, image_proj_layer in zip( + ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers + ): + output_hidden_state = not isinstance(image_proj_layer, ImageProjection) + single_image_embeds, single_negative_image_embeds = self.encode_image( + single_ip_adapter_image, device, 1, output_hidden_state + ) + + image_embeds.append(single_image_embeds[None, :]) + if do_classifier_free_guidance: + negative_image_embeds.append(single_negative_image_embeds[None, :]) + else: + for single_image_embeds in ip_adapter_image_embeds: + if do_classifier_free_guidance: + single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2) + negative_image_embeds.append(single_negative_image_embeds) + image_embeds.append(single_image_embeds) + + ip_adapter_image_embeds = [] + for i, single_image_embeds in enumerate(image_embeds): + single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0) + if do_classifier_free_guidance: + single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0) + single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0) + + single_image_embeds = single_image_embeds.to(device=device) + ip_adapter_image_embeds.append(single_image_embeds) + + return ip_adapter_image_embeds + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs + def prepare_extra_step_kwargs(self, generator, eta): + # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature + # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. + # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 + # and should be between [0, 1] + + accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) + extra_step_kwargs = {} + if accepts_eta: + extra_step_kwargs["eta"] = eta + + # check if the scheduler accepts generator + accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) + if accepts_generator: + extra_step_kwargs["generator"] = generator + return extra_step_kwargs + + def check_inputs( + self, + prompt, + image, + strength, + num_inference_steps, + callback_steps, + negative_prompt=None, + prompt_embeds=None, + negative_prompt_embeds=None, + pooled_prompt_embeds=None, + negative_pooled_prompt_embeds=None, + ip_adapter_image=None, + ip_adapter_image_embeds=None, + controlnet_conditioning_scale=1.0, + control_guidance_start=0.0, + control_guidance_end=1.0, + callback_on_step_end_tensor_inputs=None, + ): + if strength < 0 or strength > 1: + raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}") + if num_inference_steps is None: + raise ValueError("`num_inference_steps` cannot be None.") + elif not isinstance(num_inference_steps, int) or num_inference_steps <= 0: + raise ValueError( + f"`num_inference_steps` has to be a positive integer but is {num_inference_steps} of type" + f" {type(num_inference_steps)}." + ) + + if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0): + raise ValueError( + f"`callback_steps` has to be a positive integer but is {callback_steps} of type" + f" {type(callback_steps)}." + ) + + if callback_on_step_end_tensor_inputs is not None and not all( + k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs + ): + raise ValueError( + f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" + ) + + if prompt is not None and prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to" + " only forward one of the two." + ) + elif prompt is None and prompt_embeds is None: + raise ValueError( + "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." + ) + elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): + raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") + + if negative_prompt is not None and negative_prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:" + f" {negative_prompt_embeds}. Please make sure to only forward one of the two." + ) + + if prompt_embeds is not None and negative_prompt_embeds is not None: + if prompt_embeds.shape != negative_prompt_embeds.shape: + raise ValueError( + "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but" + f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`" + f" {negative_prompt_embeds.shape}." + ) + + if prompt_embeds is not None and pooled_prompt_embeds is None: + raise ValueError( + "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`." + ) + + if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None: + raise ValueError( + "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`." + ) + + # `prompt` needs more sophisticated handling when there are multiple + # conditionings. + if isinstance(self.controlnet, MultiControlNetModel): + if isinstance(prompt, list): + logger.warning( + f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}" + " prompts. The conditionings will be fixed across the prompts." + ) + + # Check `image` + is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance( + self.controlnet, torch._dynamo.eval_frame.OptimizedModule + ) + + if ( + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) + ): + self.check_image(image, prompt, prompt_embeds) + elif ( + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + ): + if not isinstance(image, list): + raise TypeError("For multiple controlnets: `image` must be type `list`") + + # When `image` is a nested list: + # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]]) + elif any(isinstance(i, list) for i in image): + raise ValueError("A single batch of multiple conditionings are supported at the moment.") + elif len(image) != len(self.controlnet.nets): + raise ValueError( + f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets." + ) + + for image_ in image: + self.check_image(image_, prompt, prompt_embeds) + else: + assert False + + # Check `controlnet_conditioning_scale` + if ( + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) + ): + if not isinstance(controlnet_conditioning_scale, float): + raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.") + elif ( + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + ): + if isinstance(controlnet_conditioning_scale, list): + if any(isinstance(i, list) for i in controlnet_conditioning_scale): + raise ValueError("A single batch of multiple conditionings are supported at the moment.") + elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len( + self.controlnet.nets + ): + raise ValueError( + "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have" + " the same length as the number of controlnets" + ) + else: + assert False + + if not isinstance(control_guidance_start, (tuple, list)): + control_guidance_start = [control_guidance_start] + + if not isinstance(control_guidance_end, (tuple, list)): + control_guidance_end = [control_guidance_end] + + if len(control_guidance_start) != len(control_guidance_end): + raise ValueError( + f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list." + ) + + if isinstance(self.controlnet, MultiControlNetModel): + if len(control_guidance_start) != len(self.controlnet.nets): + raise ValueError( + f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}." + ) + + for start, end in zip(control_guidance_start, control_guidance_end): + if start >= end: + raise ValueError( + f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}." + ) + if start < 0.0: + raise ValueError(f"control guidance start: {start} can't be smaller than 0.") + if end > 1.0: + raise ValueError(f"control guidance end: {end} can't be larger than 1.0.") + + if ip_adapter_image is not None and ip_adapter_image_embeds is not None: + raise ValueError( + "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined." + ) + + if ip_adapter_image_embeds is not None: + if not isinstance(ip_adapter_image_embeds, list): + raise ValueError( + f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}" + ) + elif ip_adapter_image_embeds[0].ndim not in [3, 4]: + raise ValueError( + f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D" + ) + + # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.check_image + def check_image(self, image, prompt, prompt_embeds): + image_is_pil = isinstance(image, PIL.Image.Image) + image_is_tensor = isinstance(image, torch.Tensor) + image_is_np = isinstance(image, np.ndarray) + image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image) + image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor) + image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray) + + if ( + not image_is_pil + and not image_is_tensor + and not image_is_np + and not image_is_pil_list + and not image_is_tensor_list + and not image_is_np_list + ): + raise TypeError( + f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}" + ) + + if image_is_pil: + image_batch_size = 1 + else: + image_batch_size = len(image) + + if prompt is not None and isinstance(prompt, str): + prompt_batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + prompt_batch_size = len(prompt) + elif prompt_embeds is not None: + prompt_batch_size = prompt_embeds.shape[0] + + if image_batch_size != 1 and image_batch_size != prompt_batch_size: + raise ValueError( + f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}" + ) + + # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.prepare_image + def prepare_control_image( + self, + image, + width, + height, + batch_size, + num_images_per_prompt, + device, + dtype, + do_classifier_free_guidance=False, + guess_mode=False, + ): + image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) + image_batch_size = image.shape[0] + + if image_batch_size == 1: + repeat_by = batch_size + else: + # image batch size is the same as prompt batch size + repeat_by = num_images_per_prompt + + image = image.repeat_interleave(repeat_by, dim=0) + + image = image.to(device=device, dtype=dtype) + + if do_classifier_free_guidance and not guess_mode: + image = torch.cat([image] * 2) + + return image + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps + def get_timesteps(self, num_inference_steps, strength, device): + # get the original timestep using init_timestep + init_timestep = min(int(num_inference_steps * strength), num_inference_steps) + + t_start = max(num_inference_steps - init_timestep, 0) + timesteps = self.scheduler.timesteps[t_start * self.scheduler.order:] + if hasattr(self.scheduler, "set_begin_index"): + self.scheduler.set_begin_index(t_start * self.scheduler.order) + + return timesteps, num_inference_steps - t_start + + # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.prepare_latents + def prepare_latents( + self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True + ): + if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)): + raise ValueError( + f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}" + ) + + # Offload text encoder if `enable_model_cpu_offload` was enabled + if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None: + torch.cuda.empty_cache() + torch.cuda.ipc_collect() + + image = image.to(device=device, dtype=dtype) + + batch_size = batch_size * num_images_per_prompt + + if image.shape[1] == 4: + init_latents = image + + else: + # make sure the VAE is in float32 mode, as it overflows in float16 + if self.vae.config.force_upcast: + image = image.float() + self.vae.to(dtype=torch.float32) + + if isinstance(generator, list) and len(generator) != batch_size: + raise ValueError( + f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" + f" size of {batch_size}. Make sure the batch size matches the length of the generators." + ) + + elif isinstance(generator, list): + init_latents = [ + retrieve_latents(self.vae.encode(image[i: i + 1]), generator=generator[i]) + for i in range(batch_size) + ] + init_latents = torch.cat(init_latents, dim=0) + else: + init_latents = retrieve_latents(self.vae.encode(image), generator=generator) + + if self.vae.config.force_upcast: + self.vae.to(dtype) + + init_latents = init_latents.to(dtype) + + init_latents = self.vae.config.scaling_factor * init_latents + + if batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] == 0: + # expand init_latents for batch_size + additional_image_per_prompt = batch_size // init_latents.shape[0] + init_latents = torch.cat([init_latents] * additional_image_per_prompt, dim=0) + elif batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] != 0: + raise ValueError( + f"Cannot duplicate `image` of batch size {init_latents.shape[0]} to {batch_size} text prompts." + ) + else: + init_latents = torch.cat([init_latents], dim=0) + + if add_noise: + shape = init_latents.shape + noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype) + # get latents + init_latents = self.scheduler.add_noise(init_latents, noise, timestep) + + latents = init_latents + + return latents + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents + def prepare_latents_t2i(self, batch_size, num_channels_latents, height, width, dtype, device, generator, + latents=None): + shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) + if isinstance(generator, list) and len(generator) != batch_size: + raise ValueError( + f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" + f" size of {batch_size}. Make sure the batch size matches the length of the generators." + ) + + if latents is None: + latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype) + else: + latents = latents.to(device) + + # scale the initial noise by the standard deviation required by the scheduler + latents = latents * self.scheduler.init_noise_sigma + return latents + + def _get_add_time_ids(self, original_size, crops_coords_top_left, target_size, dtype): + add_time_ids = list(original_size + crops_coords_top_left + target_size) + + passed_add_embed_dim = ( + self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 + ) + expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features + + if expected_add_embed_dim != passed_add_embed_dim: + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`." + ) + + add_time_ids = torch.tensor([add_time_ids], dtype=dtype) + return add_time_ids + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae + def upcast_vae(self): + dtype = self.vae.dtype + self.vae.to(dtype=torch.float32) + use_torch_2_0_or_xformers = isinstance( + self.vae.decoder.mid_block.attentions[0].processor, + ( + AttnProcessor2_0, + XFormersAttnProcessor, + ), + ) + # if xformers or torch_2_0 is used attention block does not need + # to be in float32 which can save lots of memory + if use_torch_2_0_or_xformers: + self.vae.post_quant_conv.to(dtype) + self.vae.decoder.conv_in.to(dtype) + self.vae.decoder.mid_block.to(dtype) + + @property + def guidance_scale(self): + return self._guidance_scale + + # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2) + # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1` + # corresponds to doing no classifier free guidance. + @property + def do_classifier_free_guidance(self): + return self._guidance_scale > 1 + + @property + def cross_attention_kwargs(self): + return self._cross_attention_kwargs + + @property + def num_timesteps(self): + return self._num_timesteps + + @torch.no_grad() + @replace_example_docstring(EXAMPLE_DOC_STRING) + def __call__( + self, + prompt: Union[str, List[str]] = None, + image: PipelineImageInput = None, + control_image: PipelineImageInput = None, + height: Optional[int] = None, + width: Optional[int] = None, + strength: float = 0.8, + num_inference_steps: int = 50, + guidance_scale: float = 5.0, + negative_prompt: Optional[Union[str, List[str]]] = None, + num_images_per_prompt: Optional[int] = 1, + eta: float = 0.0, + guess_mode: bool = False, + generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, + latents: Optional[torch.Tensor] = None, + prompt_embeds: Optional[torch.Tensor] = None, + negative_prompt_embeds: Optional[torch.Tensor] = None, + pooled_prompt_embeds: Optional[torch.Tensor] = None, + negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, + ip_adapter_image: Optional[PipelineImageInput] = None, + ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, + output_type: Optional[str] = "pil", + return_dict: bool = True, + cross_attention_kwargs: Optional[Dict[str, Any]] = None, + controlnet_conditioning_scale: Union[float, List[float]] = 0.8, + control_guidance_start: Union[float, List[float]] = 0.0, + control_guidance_end: Union[float, List[float]] = 1.0, + original_size: Tuple[int, int] = None, + crops_coords_top_left: Tuple[int, int] = (0, 0), + target_size: Tuple[int, int] = None, + callback_on_step_end: Optional[ + Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] + ] = None, + callback_on_step_end_tensor_inputs: List[str] = ["latents"], + **kwargs, + ): + r""" + Function invoked when calling the pipeline for generation. + + Args: + prompt (`str` or `List[str]`, *optional*): + The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`. + instead. + image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: + `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): + The initial image will be used as the starting point for the image generation process. Can also accept + image latents as `image`, if passing latents directly, it will not be encoded again. + control_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: + `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): + The ControlNet input condition. ControlNet uses this input condition to generate guidance to Unet. If + the type is specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also + be accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height + and/or width are passed, `image` is resized according to them. If multiple ControlNets are specified in + init, images must be passed as a list such that each element of the list can be correctly batched for + input to a single controlnet. + height (`int`, *optional*, defaults to the size of control_image): + The height in pixels of the generated image. Anything below 512 pixels won't work well for + [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) + and checkpoints that are not specifically fine-tuned on low resolutions. + width (`int`, *optional*, defaults to the size of control_image): + The width in pixels of the generated image. Anything below 512 pixels won't work well for + [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) + and checkpoints that are not specifically fine-tuned on low resolutions. + strength (`float`, *optional*, defaults to 0.8): + Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a + starting point and more noise is added the higher the `strength`. The number of denoising steps depends + on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising + process runs for the full number of iterations specified in `num_inference_steps`. A value of 1 + essentially ignores `image`. + num_inference_steps (`int`, *optional*, defaults to 50): + The number of denoising steps. More denoising steps usually lead to a higher quality image at the + expense of slower inference. + guidance_scale (`float`, *optional*, defaults to 7.5): + Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598). + `guidance_scale` is defined as `w` of equation 2. of [Imagen + Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale > + 1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, + usually at the expense of lower image quality. + negative_prompt (`str` or `List[str]`, *optional*): + The prompt or prompts not to guide the image generation. If not defined, one has to pass + `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is + less than `1`). + num_images_per_prompt (`int`, *optional*, defaults to 1): + The number of images to generate per prompt. + eta (`float`, *optional*, defaults to 0.0): + Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to + [`schedulers.DDIMScheduler`], will be ignored for others. + generator (`torch.Generator` or `List[torch.Generator]`, *optional*): + One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html) + to make generation deterministic. + latents (`torch.Tensor`, *optional*): + Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image + generation. Can be used to tweak the same generation with different prompts. If not provided, a latents + tensor will ge generated by sampling using the supplied random `generator`. + prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + negative_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input + argument. + pooled_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + negative_pooled_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` + input argument. + ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters. + ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*): + Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of + IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should + contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not + provided, embeddings are computed from the `ip_adapter_image` input argument. + output_type (`str`, *optional*, defaults to `"pil"`): + The output format of the generate image. Choose between + [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`. + return_dict (`bool`, *optional*, defaults to `True`): + Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a + plain tuple. + cross_attention_kwargs (`dict`, *optional*): + A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under + `self.processor` in + [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). + controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0): + The outputs of the controlnet are multiplied by `controlnet_conditioning_scale` before they are added + to the residual in the original unet. If multiple ControlNets are specified in init, you can set the + corresponding scale as a list. + control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0): + The percentage of total steps at which the controlnet starts applying. + control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0): + The percentage of total steps at which the controlnet stops applying. + original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled. + `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as + explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): + `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position + `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting + `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + For most cases, `target_size` should be set to the desired height and width of the generated image. If + not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in + section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*): + A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of + each denoising step during the inference. with the following arguments: `callback_on_step_end(self: + DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a + list of all tensors as specified by `callback_on_step_end_tensor_inputs`. + callback_on_step_end_tensor_inputs (`List`, *optional*): + The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list + will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the + `._callback_tensor_inputs` attribute of your pipeline class. + + Examples: + + Returns: + [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`: + [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple` + containing the output images. + """ + + callback = kwargs.pop("callback", None) + callback_steps = kwargs.pop("callback_steps", None) + + if callback is not None: + deprecate( + "callback", + "1.0.0", + "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", + ) + if callback_steps is not None: + deprecate( + "callback_steps", + "1.0.0", + "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", + ) + + if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)): + callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs + + controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet + + # align format for control guidance + if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list): + control_guidance_start = len(control_guidance_end) * [control_guidance_start] + elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list): + control_guidance_end = len(control_guidance_start) * [control_guidance_end] + elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list): + mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1 + control_guidance_start, control_guidance_end = ( + mult * [control_guidance_start], + mult * [control_guidance_end], + ) + + # from IPython import embed; embed() + # 1. Check inputs. Raise error if not correct + self.check_inputs( + prompt, + control_image, + strength, + num_inference_steps, + callback_steps, + negative_prompt, + prompt_embeds, + negative_prompt_embeds, + pooled_prompt_embeds, + negative_pooled_prompt_embeds, + ip_adapter_image, + ip_adapter_image_embeds, + controlnet_conditioning_scale, + control_guidance_start, + control_guidance_end, + callback_on_step_end_tensor_inputs, + ) + + self._guidance_scale = guidance_scale + self._cross_attention_kwargs = cross_attention_kwargs + + # 2. Define call parameters + if prompt is not None and isinstance(prompt, str): + batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + device = self._execution_device + + if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float): + controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets) + + # 3.1. Encode input prompt + text_encoder_lora_scale = ( + self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None + ) + ( + prompt_embeds, + negative_prompt_embeds, + pooled_prompt_embeds, + negative_pooled_prompt_embeds, + ) = self.encode_prompt( + prompt, + device, + num_images_per_prompt, + self.do_classifier_free_guidance, + negative_prompt, + prompt_embeds=prompt_embeds, + negative_prompt_embeds=negative_prompt_embeds, + pooled_prompt_embeds=pooled_prompt_embeds, + negative_pooled_prompt_embeds=negative_pooled_prompt_embeds, + lora_scale=text_encoder_lora_scale, + ) + + # 3.2 Encode ip_adapter_image + if ip_adapter_image is not None or ip_adapter_image_embeds is not None: + image_embeds = self.prepare_ip_adapter_image_embeds( + ip_adapter_image, + ip_adapter_image_embeds, + device, + batch_size * num_images_per_prompt, + self.do_classifier_free_guidance, + ) + + # 4. Prepare image and controlnet_conditioning_image + image = self.image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) + + if isinstance(controlnet, ControlNetModel): + control_image = self.prepare_control_image( + image=control_image, + width=width, + height=height, + batch_size=batch_size * num_images_per_prompt, + num_images_per_prompt=num_images_per_prompt, + device=device, + dtype=controlnet.dtype, + do_classifier_free_guidance=self.do_classifier_free_guidance, + guess_mode=guess_mode, + ) + height, width = control_image.shape[-2:] + elif isinstance(controlnet, MultiControlNetModel): + control_images = [] + + for control_image_ in control_image: + control_image_ = self.prepare_control_image( + image=control_image_, + width=width, + height=height, + batch_size=batch_size * num_images_per_prompt, + num_images_per_prompt=num_images_per_prompt, + device=device, + dtype=controlnet.dtype, + do_classifier_free_guidance=self.do_classifier_free_guidance, + guess_mode=guess_mode, + ) + + control_images.append(control_image_) + + control_image = control_images + height, width = control_image[0].shape[-2:] + else: + assert False + + # 5. Prepare timesteps + self.scheduler.set_timesteps(num_inference_steps, device=device) + timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device) + latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt) + self._num_timesteps = len(timesteps) + + # 6. Prepare latent variables + + num_channels_latents = self.unet.config.in_channels + if latents is None: + if strength >= 1.0: + latents = self.prepare_latents_t2i( + batch_size * num_images_per_prompt, + num_channels_latents, + height, + width, + prompt_embeds.dtype, + device, + generator, + latents, + ) + else: + latents = self.prepare_latents( + image, + latent_timestep, + batch_size, + num_images_per_prompt, + prompt_embeds.dtype, + device, + generator, + True, + ) + + # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline + extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta) + + # 7.1 Create tensor stating which controlnets to keep + controlnet_keep = [] + for i in range(len(timesteps)): + keeps = [ + 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e) + for s, e in zip(control_guidance_start, control_guidance_end) + ] + controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps) + + # 7.2 Prepare added time ids & embeddings + if isinstance(control_image, list): + original_size = original_size or control_image[0].shape[-2:] + else: + original_size = original_size or control_image.shape[-2:] + target_size = target_size or (height, width) + + # 7. Prepare added time ids & embeddings + add_text_embeds = pooled_prompt_embeds + add_time_ids = self._get_add_time_ids( + original_size, crops_coords_top_left, target_size, dtype=prompt_embeds.dtype + ) + + if self.do_classifier_free_guidance: + prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) + add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0) + add_time_ids = torch.cat([add_time_ids, add_time_ids], dim=0) + + prompt_embeds = prompt_embeds.to(device) + add_text_embeds = add_text_embeds.to(device) + add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1) + + # patch diffusers controlnet instance forward, undo + # after denoising loop + + patched_cn_models = [] + if isinstance(self.controlnet, MultiControlNetModel): + cn_models_to_patch = self.controlnet.nets + else: + cn_models_to_patch = [self.controlnet] + + for cn_model in cn_models_to_patch: + cn_og_forward = cn_model.forward + + def _cn_patch_forward(*args, **kwargs): + encoder_hidden_states = kwargs['encoder_hidden_states'] + if cn_model.encoder_hid_proj is not None and cn_model.config.encoder_hid_dim_type == "text_proj": + encoder_hidden_states = cn_model.encoder_hid_proj(kwargs['encoder_hidden_states']) + kwargs.pop('encoder_hidden_states') + return cn_og_forward(*args, encoder_hidden_states=encoder_hidden_states, **kwargs) + + cn_model.forward = _cn_patch_forward + patched_cn_models.append((cn_model, cn_og_forward)) + + # 8. Denoising loop + num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order + + try: + with self.progress_bar(total=num_inference_steps) as progress_bar: + for i, t in enumerate(timesteps): + # expand the latents if we are doing classifier free guidance + latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents + latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) + + added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids} + + # controlnet(s) inference + if guess_mode and self.do_classifier_free_guidance: + # Infer ControlNet only for the conditional batch. + control_model_input = latents + control_model_input = self.scheduler.scale_model_input(control_model_input, t) + controlnet_prompt_embeds = prompt_embeds.chunk(2)[1] + controlnet_added_cond_kwargs = { + "text_embeds": add_text_embeds.chunk(2)[1], + "time_ids": add_time_ids.chunk(2)[1], + } + else: + control_model_input = latent_model_input + controlnet_prompt_embeds = prompt_embeds + controlnet_added_cond_kwargs = added_cond_kwargs + + if isinstance(controlnet_keep[i], list): + cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])] + else: + controlnet_cond_scale = controlnet_conditioning_scale + if isinstance(controlnet_cond_scale, list): + controlnet_cond_scale = controlnet_cond_scale[0] + cond_scale = controlnet_cond_scale * controlnet_keep[i] + + down_block_res_samples, mid_block_res_sample = self.controlnet( + control_model_input, + t, + encoder_hidden_states=controlnet_prompt_embeds, + controlnet_cond=control_image, + conditioning_scale=cond_scale, + guess_mode=guess_mode, + added_cond_kwargs=controlnet_added_cond_kwargs, + return_dict=False, + ) + + if guess_mode and self.do_classifier_free_guidance: + # Infered ControlNet only for the conditional batch. + # To apply the output of ControlNet to both the unconditional and conditional batches, + # add 0 to the unconditional batch to keep it unchanged. + down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples] + mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample]) + + if ip_adapter_image is not None or ip_adapter_image_embeds is not None: + added_cond_kwargs["image_embeds"] = image_embeds + + # predict the noise residual + noise_pred = self.unet( + latent_model_input, + t, + encoder_hidden_states=prompt_embeds, + cross_attention_kwargs=self.cross_attention_kwargs, + down_block_additional_residuals=down_block_res_samples, + mid_block_additional_residual=mid_block_res_sample, + added_cond_kwargs=added_cond_kwargs, + return_dict=False, + )[0] + + # perform guidance + if self.do_classifier_free_guidance: + noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) + noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) + + # compute the previous noisy sample x_t -> x_t-1 + latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] + + # call the callback, if provided + if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): + progress_bar.update() + if callback is not None and i % callback_steps == 0: + step_idx = i // getattr(self.scheduler, "order", 1) + callback(step_idx, t, latents) + finally: + for cn_and_og in patched_cn_models: + cn_and_og[0].forward = cn_and_og[1] + + # If we do sequential model offloading, let's offload unet and controlnet + # manually for max memory savings + if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None: + self.unet.to("cpu") + self.controlnet.to("cpu") + torch.cuda.empty_cache() + torch.cuda.ipc_collect() + + if not output_type == "latent": + # make sure the VAE is in float32 mode, as it overflows in float16 + needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast + + if needs_upcasting: + self.upcast_vae() + latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype) + + latents = latents / self.vae.config.scaling_factor + image = self.vae.decode(latents, return_dict=False)[0] + + # cast back to fp16 if needed + if needs_upcasting: + self.vae.to(dtype=torch.float16) + else: + image = latents + return StableDiffusionXLPipelineOutput(images=image) + + image = self.image_processor.postprocess(image, output_type=output_type) + + # Offload all models + self.maybe_free_model_hooks() + + if not return_dict: + return (image,) + + return StableDiffusionXLPipelineOutput(images=image) diff --git a/examples/community/pipeline_controlnet_xl_kolors_inpaint.py b/examples/community/pipeline_controlnet_xl_kolors_inpaint.py new file mode 100644 index 000000000000..e587b0f3dcd2 --- /dev/null +++ b/examples/community/pipeline_controlnet_xl_kolors_inpaint.py @@ -0,0 +1,1807 @@ +# Copyright 2024 The HuggingFace Team. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import inspect +from typing import Any, Callable, Dict, List, Optional, Tuple, Union + +import PIL.Image +import numpy as np +import torch +import torch.nn.functional as F +from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback +from diffusers.image_processor import PipelineImageInput, VaeImageProcessor +from diffusers.loaders import ( + FromSingleFileMixin, + IPAdapterMixin, + StableDiffusionXLLoraLoaderMixin, + TextualInversionLoaderMixin +) +from diffusers.models import AutoencoderKL, ImageProjection, UNet2DConditionModel +from diffusers.models import ControlNetModel, MultiControlNetModel +from diffusers.models.attention_processor import ( + AttnProcessor2_0, + XFormersAttnProcessor, +) + +from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin +from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput +from diffusers.schedulers import KarrasDiffusionSchedulers +from diffusers.utils import ( + is_invisible_watermark_available, + replace_example_docstring, + deprecate, + logging +) +from diffusers.utils.torch_utils import is_compiled_module, randn_tensor +from transformers import ( + CLIPImageProcessor, + CLIPVisionModelWithProjection, +) + +from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer + +if is_invisible_watermark_available(): + from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker + +logger = logging.get_logger(__name__) # pylint: disable=invalid-name + + +EXAMPLE_DOC_STRING = """ + Examples: + ```py + >>> from diffusers import KolorsControlNetInpaintPipeline, ControlNetModel + >>> from diffusers.utils import load_image + >>> from PIL import Image + >>> import numpy as np + >>> import torch + >>> import cv2 + + >>> init_image = load_image( + ... "https://huggingface.co/datasets/diffusers/test-arrays/resolve/main/stable_diffusion_inpaint/boy.png" + ... ) + >>> init_image = init_image.resize((1024, 1024)) + + >>> generator = torch.Generator(device="cpu").manual_seed(1) + + >>> mask_image = load_image( + ... "https://huggingface.co/datasets/diffusers/test-arrays/resolve/main/stable_diffusion_inpaint/boy_mask.png" + ... ) + >>> mask_image = mask_image.resize((1024, 1024)) + + + >>> def make_canny_condition(image): + ... image = np.array(image) + ... image = cv2.Canny(image, 100, 200) + ... image = image[:, :, None] + ... image = np.concatenate([image, image, image], axis=2) + ... image = Image.fromarray(image) + ... return image + + + >>> control_image = make_canny_condition(init_image) + + >>> controlnet = ControlNetModel.from_pretrained( + ... "Kwai-Kolors/Kolors-ControlNet-Canny", torch_dtype=torch.float16 + ... ) + >>> pipe = KolorsControlNetInpaintPipeline.from_pretrained( + ... "Kwai-Kolors/Kolors-diffusers", controlnet=controlnet, torch_dtype=torch.float16 + ... ) + + >>> pipe.enable_model_cpu_offload() + + # generate image + >>> image = pipe( + ... "a handsome man with ray-ban sunglasses", + ... num_inference_steps=20, + ... generator=generator, + ... eta=1.0, + ... image=init_image, + ... mask_image=mask_image, + ... control_image=control_image, + ... ).images[0] + ``` +""" + + +# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents +def retrieve_latents( + encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" +): + if hasattr(encoder_output, "latent_dist") and sample_mode == "sample": + return encoder_output.latent_dist.sample(generator) + elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax": + return encoder_output.latent_dist.mode() + elif hasattr(encoder_output, "latents"): + return encoder_output.latents + else: + raise AttributeError("Could not access latents of provided encoder_output") + + +# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps +def retrieve_timesteps( + scheduler, + num_inference_steps: Optional[int] = None, + device: Optional[Union[str, torch.device]] = None, + timesteps: Optional[List[int]] = None, + sigmas: Optional[List[float]] = None, + **kwargs, +): + """ + Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles + custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. + + Args: + scheduler (`SchedulerMixin`): + The scheduler to get timesteps from. + num_inference_steps (`int`): + The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps` + must be `None`. + device (`str` or `torch.device`, *optional*): + The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. + timesteps (`List[int]`, *optional*): + Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed, + `num_inference_steps` and `sigmas` must be `None`. + sigmas (`List[float]`, *optional*): + Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed, + `num_inference_steps` and `timesteps` must be `None`. + + Returns: + `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the + second element is the number of inference steps. + """ + if timesteps is not None and sigmas is not None: + raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values") + if timesteps is not None: + accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) + if not accepts_timesteps: + raise ValueError( + f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" + f" timestep schedules. Please check whether you are using the correct scheduler." + ) + scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs) + timesteps = scheduler.timesteps + num_inference_steps = len(timesteps) + elif sigmas is not None: + accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) + if not accept_sigmas: + raise ValueError( + f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" + f" sigmas schedules. Please check whether you are using the correct scheduler." + ) + scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs) + timesteps = scheduler.timesteps + num_inference_steps = len(timesteps) + else: + scheduler.set_timesteps(num_inference_steps, device=device, **kwargs) + timesteps = scheduler.timesteps + return timesteps, num_inference_steps + + +class KolorsControlNetInpaintPipeline( + DiffusionPipeline, + StableDiffusionMixin, + StableDiffusionXLLoraLoaderMixin, + FromSingleFileMixin, + IPAdapterMixin, +): + r""" + Pipeline for inpainting using Kolors with ControlNet guidance. + + This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the + library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.) + + The pipeline also inherits the following loading methods: + - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.safetensors` files + - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights + - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights + - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters + + Args: + vae ([`AutoencoderKL`]): + Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. + text_encoder ([`ChatGLMModel`]): + Frozen text-encoder. Kolors uses [ChatGLM3-6B](https://huggingface.co/THUDM/chatglm3-6b). + tokenizer (`ChatGLMTokenizer`): + Tokenizer of class + [ChatGLMTokenizer](https://huggingface.co/THUDM/chatglm3-6b/blob/main/tokenization_chatglm.py). + unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents. + controlnet ([`ControlNetModel`] or `List[ControlNetModel]`): + Provides additional conditioning to the unet during the denoising process. If you set multiple ControlNets + as a list, the outputs from each ControlNet are added together to create one combined additional + conditioning. + scheduler ([`SchedulerMixin`]): + A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of + [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`]. + requires_aesthetics_score (`bool`, *optional*, defaults to `"False"`): + Whether the `unet` requires an `aesthetic_score` condition to be passed during inference. Also see the + config of `stabilityai/stable-diffusion-xl-refiner-1-0`. + force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`): + Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of + `Kwai-Kolors/Kolors-diffusers`. + feature_extractor ([`~transformers.CLIPImageProcessor`]): + A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`. + """ + + model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae" + + _optional_components = [ + "tokenizer", + "text_encoder", + "feature_extractor", + "image_encoder", + ] + _callback_tensor_inputs = [ + "latents", + "prompt_embeds", + "negative_prompt_embeds", + "add_text_embeds", + "add_time_ids", + "negative_pooled_prompt_embeds", + "add_neg_time_ids", + ] + + def __init__( + self, + vae: AutoencoderKL, + text_encoder: ChatGLMModel, + tokenizer: ChatGLMTokenizer, + unet: UNet2DConditionModel, + controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel], + scheduler: KarrasDiffusionSchedulers, + requires_aesthetics_score: bool = False, + force_zeros_for_empty_prompt: bool = True, + feature_extractor: CLIPImageProcessor = None, + image_encoder: CLIPVisionModelWithProjection = None, + add_watermarker: Optional[bool] = None, + ): + super().__init__() + + if isinstance(controlnet, (list, tuple)): + controlnet = MultiControlNetModel(controlnet) + + self.register_modules( + vae=vae, + text_encoder=text_encoder, + tokenizer=tokenizer, + unet=unet, + controlnet=controlnet, + scheduler=scheduler, + feature_extractor=feature_extractor, + image_encoder=image_encoder, + ) + self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) + self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True) + self.control_image_processor = VaeImageProcessor( + vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False + ) + self.mask_processor = VaeImageProcessor( + vae_scale_factor=self.vae_scale_factor, do_normalize=False, do_binarize=True, do_convert_grayscale=True + ) + + if add_watermarker: + self.watermark = StableDiffusionXLWatermarker() + else: + self.watermark = None + + self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt) + self.register_to_config(requires_aesthetics_score=requires_aesthetics_score) + + def encode_prompt( + self, + prompt, + device: Optional[torch.device] = None, + num_images_per_prompt: int = 1, + do_classifier_free_guidance: bool = True, + negative_prompt=None, + prompt_embeds: Optional[torch.FloatTensor] = None, + negative_prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + lora_scale: Optional[float] = None, + ): + r""" + Encodes the prompt into text encoder hidden states. + + Args: + prompt (`str` or `List[str]`, *optional*): + prompt to be encoded + device: (`torch.device`): + torch device + num_images_per_prompt (`int`): + number of images that should be generated per prompt + do_classifier_free_guidance (`bool`): + whether to use classifier free guidance or not + negative_prompt (`str` or `List[str]`, *optional*): + The prompt or prompts not to guide the image generation. If not defined, one has to pass + `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is + less than `1`). + prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + negative_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input + argument. + pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` + input argument. + lora_scale (`float`, *optional*): + A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. + """ + device = device or self._execution_device + + # set lora scale so that monkey patched LoRA + # function of text encoder can correctly access it + if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin): + self._lora_scale = lora_scale + + if prompt is not None and isinstance(prompt, str): + batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + # Define tokenizers and text encoders + tokenizers = [self.tokenizer] + text_encoders = [self.text_encoder] + + if prompt_embeds is None: + # textual inversion: procecss multi-vector tokens if necessary + prompt_embeds_list = [] + for tokenizer, text_encoder in zip(tokenizers, text_encoders): + if isinstance(self, TextualInversionLoaderMixin): + prompt = self.maybe_convert_prompt(prompt, tokenizer) + + text_inputs = tokenizer( + prompt, + padding="max_length", + max_length=256, + truncation=True, + return_tensors="pt", + ).to(self._execution_device) + output = text_encoder( + input_ids=text_inputs['input_ids'], + attention_mask=text_inputs['attention_mask'], + position_ids=text_inputs['position_ids'], + output_hidden_states=True) + prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() + pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] + bs_embed, seq_len, _ = prompt_embeds.shape + prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) + prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1) + prompt_embeds_list.append(prompt_embeds) + + # prompt_embeds = torch.concat(prompt_embeds_list, dim=-1) + prompt_embeds = prompt_embeds_list[0] + + # get unconditional embeddings for classifier free guidance + zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt + if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt: + negative_prompt_embeds = torch.zeros_like(prompt_embeds) + negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds) + elif do_classifier_free_guidance and negative_prompt_embeds is None: + # negative_prompt = negative_prompt or "" + uncond_tokens: List[str] + if negative_prompt is None: + uncond_tokens = [""] * batch_size + elif prompt is not None and type(prompt) is not type(negative_prompt): + raise TypeError( + f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" + f" {type(prompt)}." + ) + elif isinstance(negative_prompt, str): + uncond_tokens = [negative_prompt] + elif batch_size != len(negative_prompt): + raise ValueError( + f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" + f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" + " the batch size of `prompt`." + ) + else: + uncond_tokens = negative_prompt + + negative_prompt_embeds_list = [] + for tokenizer, text_encoder in zip(tokenizers, text_encoders): + # textual inversion: procecss multi-vector tokens if necessary + if isinstance(self, TextualInversionLoaderMixin): + uncond_tokens = self.maybe_convert_prompt(uncond_tokens, tokenizer) + + max_length = prompt_embeds.shape[1] + uncond_input = tokenizer( + uncond_tokens, + padding="max_length", + max_length=max_length, + truncation=True, + return_tensors="pt", + ).to(self._execution_device) + output = text_encoder( + input_ids=uncond_input['input_ids'], + attention_mask=uncond_input['attention_mask'], + position_ids=uncond_input['position_ids'], + output_hidden_states=True) + negative_prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() + negative_pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] + + if do_classifier_free_guidance: + # duplicate unconditional embeddings for each generation per prompt, using mps friendly method + seq_len = negative_prompt_embeds.shape[1] + + negative_prompt_embeds = negative_prompt_embeds.to(dtype=text_encoder.dtype, device=device) + + negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1) + negative_prompt_embeds = negative_prompt_embeds.view( + batch_size * num_images_per_prompt, seq_len, -1 + ) + + # For classifier free guidance, we need to do two forward passes. + # Here we concatenate the unconditional and text embeddings into a single batch + # to avoid doing two forward passes + + negative_prompt_embeds_list.append(negative_prompt_embeds) + + # negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1) + negative_prompt_embeds = negative_prompt_embeds_list[0] + + bs_embed = pooled_prompt_embeds.shape[0] + pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( + bs_embed * num_images_per_prompt, -1 + ) + if do_classifier_free_guidance: + negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( + bs_embed * num_images_per_prompt, -1 + ) + + return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image + def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None): + dtype = next(self.image_encoder.parameters()).dtype + + if not isinstance(image, torch.Tensor): + image = self.feature_extractor(image, return_tensors="pt").pixel_values + + image = image.to(device=device, dtype=dtype) + if output_hidden_states: + image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2] + image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_enc_hidden_states = self.image_encoder( + torch.zeros_like(image), output_hidden_states=True + ).hidden_states[-2] + uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave( + num_images_per_prompt, dim=0 + ) + return image_enc_hidden_states, uncond_image_enc_hidden_states + else: + image_embeds = self.image_encoder(image).image_embeds + image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_embeds = torch.zeros_like(image_embeds) + + return image_embeds, uncond_image_embeds + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds + def prepare_ip_adapter_image_embeds( + self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance + ): + image_embeds = [] + if do_classifier_free_guidance: + negative_image_embeds = [] + if ip_adapter_image_embeds is None: + if not isinstance(ip_adapter_image, list): + ip_adapter_image = [ip_adapter_image] + + if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers): + raise ValueError( + f"`ip_adapter_image` must have same length as the number of IP Adapters. Got " + f"{len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters." + ) + + for single_ip_adapter_image, image_proj_layer in zip( + ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers + ): + output_hidden_state = not isinstance(image_proj_layer, ImageProjection) + single_image_embeds, single_negative_image_embeds = self.encode_image( + single_ip_adapter_image, device, 1, output_hidden_state + ) + + image_embeds.append(single_image_embeds[None, :]) + if do_classifier_free_guidance: + negative_image_embeds.append(single_negative_image_embeds[None, :]) + else: + for single_image_embeds in ip_adapter_image_embeds: + if do_classifier_free_guidance: + single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2) + negative_image_embeds.append(single_negative_image_embeds) + image_embeds.append(single_image_embeds) + + ip_adapter_image_embeds = [] + for i, single_image_embeds in enumerate(image_embeds): + single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0) + if do_classifier_free_guidance: + single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0) + single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0) + + single_image_embeds = single_image_embeds.to(device=device) + ip_adapter_image_embeds.append(single_image_embeds) + + return ip_adapter_image_embeds + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs + def prepare_extra_step_kwargs(self, generator, eta): + # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature + # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. + # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 + # and should be between [0, 1] + + accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) + extra_step_kwargs = {} + if accepts_eta: + extra_step_kwargs["eta"] = eta + + # check if the scheduler accepts generator + accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) + if accepts_generator: + extra_step_kwargs["generator"] = generator + return extra_step_kwargs + + def check_inputs( + self, + prompt, + image, + strength, + num_inference_steps, + callback_steps, + negative_prompt=None, + prompt_embeds=None, + negative_prompt_embeds=None, + pooled_prompt_embeds=None, + negative_pooled_prompt_embeds=None, + ip_adapter_image=None, + ip_adapter_image_embeds=None, + controlnet_conditioning_scale=1.0, + control_guidance_start=0.0, + control_guidance_end=1.0, + callback_on_step_end_tensor_inputs=None, + ): + if strength < 0 or strength > 1: + raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}") + if num_inference_steps is None: + raise ValueError("`num_inference_steps` cannot be None.") + elif not isinstance(num_inference_steps, int) or num_inference_steps <= 0: + raise ValueError( + f"`num_inference_steps` has to be a positive integer but is {num_inference_steps} of type" + f" {type(num_inference_steps)}." + ) + + if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0): + raise ValueError( + f"`callback_steps` has to be a positive integer but is {callback_steps} of type" + f" {type(callback_steps)}." + ) + + if callback_on_step_end_tensor_inputs is not None and not all( + k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs + ): + raise ValueError( + f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" + ) + + if prompt is not None and prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to" + " only forward one of the two." + ) + elif prompt is None and prompt_embeds is None: + raise ValueError( + "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." + ) + elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): + raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") + + if negative_prompt is not None and negative_prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:" + f" {negative_prompt_embeds}. Please make sure to only forward one of the two." + ) + + if prompt_embeds is not None and negative_prompt_embeds is not None: + if prompt_embeds.shape != negative_prompt_embeds.shape: + raise ValueError( + "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but" + f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`" + f" {negative_prompt_embeds.shape}." + ) + + if prompt_embeds is not None and pooled_prompt_embeds is None: + raise ValueError( + "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`." + ) + + if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None: + raise ValueError( + "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`." + ) + + # `prompt` needs more sophisticated handling when there are multiple + # conditionings. + if isinstance(self.controlnet, MultiControlNetModel): + if isinstance(prompt, list): + logger.warning( + f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}" + " prompts. The conditionings will be fixed across the prompts." + ) + + # Check `image` + is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance( + self.controlnet, torch._dynamo.eval_frame.OptimizedModule + ) + + if ( + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) + ): + self.check_image(image, prompt, prompt_embeds) + elif ( + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + ): + if not isinstance(image, list): + raise TypeError("For multiple controlnets: `image` must be type `list`") + + # When `image` is a nested list: + # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]]) + elif any(isinstance(i, list) for i in image): + raise ValueError("A single batch of multiple conditionings are supported at the moment.") + elif len(image) != len(self.controlnet.nets): + raise ValueError( + f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets." + ) + + for image_ in image: + self.check_image(image_, prompt, prompt_embeds) + else: + assert False + + # Check `controlnet_conditioning_scale` + if ( + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) + ): + if not isinstance(controlnet_conditioning_scale, float): + raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.") + elif ( + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + ): + if isinstance(controlnet_conditioning_scale, list): + if any(isinstance(i, list) for i in controlnet_conditioning_scale): + raise ValueError("A single batch of multiple conditionings are supported at the moment.") + elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len( + self.controlnet.nets + ): + raise ValueError( + "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have" + " the same length as the number of controlnets" + ) + else: + assert False + + if not isinstance(control_guidance_start, (tuple, list)): + control_guidance_start = [control_guidance_start] + + if not isinstance(control_guidance_end, (tuple, list)): + control_guidance_end = [control_guidance_end] + + if len(control_guidance_start) != len(control_guidance_end): + raise ValueError( + f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list." + ) + + if isinstance(self.controlnet, MultiControlNetModel): + if len(control_guidance_start) != len(self.controlnet.nets): + raise ValueError( + f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}." + ) + + for start, end in zip(control_guidance_start, control_guidance_end): + if start >= end: + raise ValueError( + f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}." + ) + if start < 0.0: + raise ValueError(f"control guidance start: {start} can't be smaller than 0.") + if end > 1.0: + raise ValueError(f"control guidance end: {end} can't be larger than 1.0.") + + if ip_adapter_image is not None and ip_adapter_image_embeds is not None: + raise ValueError( + "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined." + ) + + if ip_adapter_image_embeds is not None: + if not isinstance(ip_adapter_image_embeds, list): + raise ValueError( + f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}" + ) + elif ip_adapter_image_embeds[0].ndim not in [3, 4]: + raise ValueError( + f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D" + ) + + # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.check_image + def check_image(self, image, prompt, prompt_embeds): + image_is_pil = isinstance(image, PIL.Image.Image) + image_is_tensor = isinstance(image, torch.Tensor) + image_is_np = isinstance(image, np.ndarray) + image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image) + image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor) + image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray) + + if ( + not image_is_pil + and not image_is_tensor + and not image_is_np + and not image_is_pil_list + and not image_is_tensor_list + and not image_is_np_list + ): + raise TypeError( + f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}" + ) + + if image_is_pil: + image_batch_size = 1 + else: + image_batch_size = len(image) + + if prompt is not None and isinstance(prompt, str): + prompt_batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + prompt_batch_size = len(prompt) + elif prompt_embeds is not None: + prompt_batch_size = prompt_embeds.shape[0] + + if image_batch_size != 1 and image_batch_size != prompt_batch_size: + raise ValueError( + f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}" + ) + + # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.prepare_image + def prepare_control_image( + self, + image, + width, + height, + batch_size, + num_images_per_prompt, + device, + dtype, + do_classifier_free_guidance=False, + guess_mode=False, + ): + image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) + image_batch_size = image.shape[0] + + if image_batch_size == 1: + repeat_by = batch_size + else: + # image batch size is the same as prompt batch size + repeat_by = num_images_per_prompt + + image = image.repeat_interleave(repeat_by, dim=0) + + image = image.to(device=device, dtype=dtype) + + if do_classifier_free_guidance and not guess_mode: + image = torch.cat([image] * 2) + + return image + + # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.get_timesteps + def get_timesteps(self, num_inference_steps, strength, device, denoising_start=None): + # get the original timestep using init_timestep + if denoising_start is None: + init_timestep = min(int(num_inference_steps * strength), num_inference_steps) + t_start = max(num_inference_steps - init_timestep, 0) + else: + t_start = 0 + + timesteps = self.scheduler.timesteps[t_start * self.scheduler.order:] + + # Strength is irrelevant if we directly request a timestep to start at; + # that is, strength is determined by the denoising_start instead. + if denoising_start is not None: + discrete_timestep_cutoff = int( + round( + self.scheduler.config.num_train_timesteps + - (denoising_start * self.scheduler.config.num_train_timesteps) + ) + ) + + num_inference_steps = (timesteps < discrete_timestep_cutoff).sum().item() + if self.scheduler.order == 2 and num_inference_steps % 2 == 0: + # if the scheduler is a 2nd order scheduler we might have to do +1 + # because `num_inference_steps` might be even given that every timestep + # (except the highest one) is duplicated. If `num_inference_steps` is even it would + # mean that we cut the timesteps in the middle of the denoising step + # (between 1st and 2nd derivative) which leads to incorrect results. By adding 1 + # we ensure that the denoising process always ends after the 2nd derivate step of the scheduler + num_inference_steps = num_inference_steps + 1 + + # because t_n+1 >= t_n, we slice the timesteps starting from the end + timesteps = timesteps[-num_inference_steps:] + return timesteps, num_inference_steps + + return timesteps, num_inference_steps - t_start + + # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.prepare_latents + def prepare_latents( + self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True + ): + if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)): + raise ValueError( + f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}" + ) + + # Offload text encoder if `enable_model_cpu_offload` was enabled + if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None: + torch.cuda.empty_cache() + torch.cuda.ipc_collect() + + image = image.to(device=device, dtype=dtype) + + batch_size = batch_size * num_images_per_prompt + + if image.shape[1] == 4: + init_latents = image + + else: + # make sure the VAE is in float32 mode, as it overflows in float16 + if self.vae.config.force_upcast: + image = image.float() + self.vae.to(dtype=torch.float32) + + if isinstance(generator, list) and len(generator) != batch_size: + raise ValueError( + f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" + f" size of {batch_size}. Make sure the batch size matches the length of the generators." + ) + + elif isinstance(generator, list): + init_latents = [ + retrieve_latents(self.vae.encode(image[i: i + 1]), generator=generator[i]) + for i in range(batch_size) + ] + init_latents = torch.cat(init_latents, dim=0) + else: + init_latents = retrieve_latents(self.vae.encode(image), generator=generator) + + if self.vae.config.force_upcast: + self.vae.to(dtype) + + init_latents = init_latents.to(dtype) + + init_latents = self.vae.config.scaling_factor * init_latents + + if batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] == 0: + # expand init_latents for batch_size + additional_image_per_prompt = batch_size // init_latents.shape[0] + init_latents = torch.cat([init_latents] * additional_image_per_prompt, dim=0) + elif batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] != 0: + raise ValueError( + f"Cannot duplicate `image` of batch size {init_latents.shape[0]} to {batch_size} text prompts." + ) + else: + init_latents = torch.cat([init_latents], dim=0) + + if add_noise: + shape = init_latents.shape + noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype) + # get latents + init_latents = self.scheduler.add_noise(init_latents, noise, timestep) + + latents = init_latents + + return latents + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents + def prepare_latents_t2i(self, batch_size, num_channels_latents, height, width, dtype, device, generator, + latents=None): + shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) + if isinstance(generator, list) and len(generator) != batch_size: + raise ValueError( + f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" + f" size of {batch_size}. Make sure the batch size matches the length of the generators." + ) + + if latents is None: + latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype) + else: + latents = latents.to(device) + + # scale the initial noise by the standard deviation required by the scheduler + latents = latents * self.scheduler.init_noise_sigma + return latents + + def _get_add_time_ids(self, original_size, crops_coords_top_left, target_size, dtype): + add_time_ids = list(original_size + crops_coords_top_left + target_size) + + passed_add_embed_dim = ( + self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 + ) + expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features + + if expected_add_embed_dim != passed_add_embed_dim: + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector " + f"of {passed_add_embed_dim} was created. The model has an incorrect config. Please check " + f"`unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`." + ) + + add_time_ids = torch.tensor([add_time_ids], dtype=dtype) + return add_time_ids + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae + def upcast_vae(self): + dtype = self.vae.dtype + self.vae.to(dtype=torch.float32) + use_torch_2_0_or_xformers = isinstance( + self.vae.decoder.mid_block.attentions[0].processor, + ( + AttnProcessor2_0, + XFormersAttnProcessor, + ), + ) + # if xformers or torch_2_0 is used attention block does not need + # to be in float32 which can save lots of memory + if use_torch_2_0_or_xformers: + self.vae.post_quant_conv.to(dtype) + self.vae.decoder.conv_in.to(dtype) + self.vae.decoder.mid_block.to(dtype) + + @property + def denoising_end(self): + return self._denoising_end + + @property + def denoising_start(self): + return self._denoising_start + + @property + def guidance_scale(self): + return self._guidance_scale + + # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2) + # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1` + # corresponds to doing no classifier free guidance. + @property + def do_classifier_free_guidance(self): + return self._guidance_scale > 1 + + @property + def cross_attention_kwargs(self): + return self._cross_attention_kwargs + + @property + def num_timesteps(self): + return self._num_timesteps + + def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator): + dtype = image.dtype + if self.vae.config.force_upcast: + image = image.float() + self.vae.to(dtype=torch.float32) + + if isinstance(generator, list): + image_latents = [ + retrieve_latents(self.vae.encode(image[i: i + 1]), generator=generator[i]) + for i in range(image.shape[0]) + ] + image_latents = torch.cat(image_latents, dim=0) + else: + image_latents = retrieve_latents(self.vae.encode(image), generator=generator) + + if self.vae.config.force_upcast: + self.vae.to(dtype) + + image_latents = image_latents.to(dtype) + image_latents = self.vae.config.scaling_factor * image_latents + + return image_latents + + def prepare_mask_latents( + self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance + ): + # resize the mask to latents shape as we concatenate the mask to the latents + # we do that before converting to dtype to avoid breaking in case we're using cpu_offload + # and half precision + mask = torch.nn.functional.interpolate( + mask, size=(height // self.vae_scale_factor, width // self.vae_scale_factor) + ) + mask = mask.to(device=device, dtype=dtype) + + # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method + if mask.shape[0] < batch_size: + if not batch_size % mask.shape[0] == 0: + raise ValueError( + "The passed mask and the required batch size don't match. Masks are supposed to be duplicated to" + f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number" + " of masks that you pass is divisible by the total requested batch size." + ) + mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1) + + mask = torch.cat([mask] * 2) if do_classifier_free_guidance else mask + + if masked_image is not None and masked_image.shape[1] == 4: + masked_image_latents = masked_image + else: + masked_image_latents = None + + if masked_image is not None: + if masked_image_latents is None: + masked_image = masked_image.to(device=device, dtype=dtype) + masked_image_latents = self._encode_vae_image(masked_image, generator=generator) + + if masked_image_latents.shape[0] < batch_size: + if not batch_size % masked_image_latents.shape[0] == 0: + raise ValueError( + "The passed images and the required batch size don't match. Images are supposed to be duplicated" + f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed." + " Make sure the number of images that you pass is divisible by the total requested batch size." + ) + masked_image_latents = masked_image_latents.repeat( + batch_size // masked_image_latents.shape[0], 1, 1, 1 + ) + + masked_image_latents = ( + torch.cat([masked_image_latents] * 2) if do_classifier_free_guidance else masked_image_latents + ) + + # aligning device to prevent device errors when concating it with the latent model input + masked_image_latents = masked_image_latents.to(device=device, dtype=dtype) + + return mask, masked_image_latents + + @torch.no_grad() + @replace_example_docstring(EXAMPLE_DOC_STRING) + def __call__( + self, + prompt: Union[str, List[str]] = None, + image: PipelineImageInput = None, + mask_image: PipelineImageInput = None, + control_image: PipelineImageInput = None, + masked_image_latents: torch.Tensor = None, + height: Optional[int] = None, + width: Optional[int] = None, + padding_mask_crop: Optional[int] = None, + strength: float = 0.9999, + num_inference_steps: int = 50, + timesteps: List[int] = None, + sigmas: List[float] = None, + denoising_start: Optional[float] = None, + denoising_end: Optional[float] = None, + guidance_scale: float = 7.5, + negative_prompt: Optional[Union[str, List[str]]] = None, + num_images_per_prompt: Optional[int] = 1, + eta: float = 0.0, + guess_mode: bool = False, + generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, + latents: Optional[torch.Tensor] = None, + prompt_embeds: Optional[torch.Tensor] = None, + negative_prompt_embeds: Optional[torch.Tensor] = None, + pooled_prompt_embeds: Optional[torch.Tensor] = None, + negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, + ip_adapter_image: Optional[PipelineImageInput] = None, + ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, + output_type: Optional[str] = "pil", + return_dict: bool = True, + cross_attention_kwargs: Optional[Dict[str, Any]] = None, + controlnet_conditioning_scale: Union[float, List[float]] = 0.8, + control_guidance_start: Union[float, List[float]] = 0.0, + control_guidance_end: Union[float, List[float]] = 1.0, + guidance_rescale: float = 0.0, + original_size: Tuple[int, int] = None, + crops_coords_top_left: Tuple[int, int] = (0, 0), + target_size: Tuple[int, int] = None, + negative_original_size: Optional[Tuple[int, int]] = None, + negative_crops_coords_top_left: Tuple[int, int] = (0, 0), + negative_target_size: Optional[Tuple[int, int]] = None, + aesthetic_score: float = 6.0, + negative_aesthetic_score: float = 2.5, + callback_on_step_end: Optional[ + Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] + ] = None, + callback_on_step_end_tensor_inputs: List[str] = ["latents"], + **kwargs, + ): + r""" + Function invoked when calling the pipeline for generation. + + Args: + prompt (`str` or `List[str]`, *optional*): + The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`. + instead. + prompt_2 (`str` or `List[str]`, *optional*): + The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is + used in both text-encoders + image (`PIL.Image.Image`): + `Image`, or tensor representing an image batch which will be inpainted, *i.e.* parts of the image will + be masked out with `mask_image` and repainted according to `prompt`. + mask_image (`PIL.Image.Image`): + `Image`, or tensor representing an image batch, to mask `image`. White pixels in the mask will be + repainted, while black pixels will be preserved. If `mask_image` is a PIL image, it will be converted + to a single channel (luminance) before use. If it's a tensor, it should contain one color channel (L) + instead of 3, so the expected shape would be `(B, H, W, 1)`. + control_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: + `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): + The ControlNet input condition. ControlNet uses this input condition to generate guidance to Unet. If + the type is specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also + be accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height + and/or width are passed, `image` is resized according to them. If multiple ControlNets are specified in + init, images must be passed as a list such that each element of the list can be correctly batched for + input to a single controlnet. + height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor): + The height in pixels of the generated image. This is set to 1024 by default for the best results. + Anything below 512 pixels won't work well for + [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) + and checkpoints that are not specifically fine-tuned on low resolutions. + width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor): + The width in pixels of the generated image. This is set to 1024 by default for the best results. + Anything below 512 pixels won't work well for + [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) + and checkpoints that are not specifically fine-tuned on low resolutions. + padding_mask_crop (`int`, *optional*, defaults to `None`): + The size of margin in the crop to be applied to the image and masking. If `None`, no crop is applied to + image and mask_image. If `padding_mask_crop` is not `None`, it will first find a rectangular region + with the same aspect ration of the image and contains all masked area, and then expand that area based + on `padding_mask_crop`. The image and mask_image will then be cropped based on the expanded area before + resizing to the original image size for inpainting. This is useful when the masked area is small while + the image is large and contain information irrelevant for inpainting, such as background. + strength (`float`, *optional*, defaults to 0.9999): + Conceptually, indicates how much to transform the masked portion of the reference `image`. Must be + between 0 and 1. `image` will be used as a starting point, adding more noise to it the larger the + `strength`. The number of denoising steps depends on the amount of noise initially added. When + `strength` is 1, added noise will be maximum and the denoising process will run for the full number of + iterations specified in `num_inference_steps`. A value of 1, therefore, essentially ignores the masked + portion of the reference `image`. Note that in the case of `denoising_start` being declared as an + integer, the value of `strength` will be ignored. + num_inference_steps (`int`, *optional*, defaults to 50): + The number of denoising steps. More denoising steps usually lead to a higher quality image at the + expense of slower inference. + timesteps (`List[int]`, *optional*): + Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument + in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is + passed will be used. Must be in descending order. + sigmas (`List[float]`, *optional*): + Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in + their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed + will be used. + denoising_start (`float`, *optional*): + When specified, indicates the fraction (between 0.0 and 1.0) of the total denoising process to be + bypassed before it is initiated. Consequently, the initial part of the denoising process is skipped and + it is assumed that the passed `image` is a partly denoised image. Note that when this is specified, + strength will be ignored. The `denoising_start` parameter is particularly beneficial when this pipeline + is integrated into a "Mixture of Denoisers" multi-pipeline setup, as detailed in [**Refining the Image + Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output). + denoising_end (`float`, *optional*): + When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be + completed before it is intentionally prematurely terminated. As a result, the returned sample will + still retain a substantial amount of noise (ca. final 20% of timesteps still needed) and should be + denoised by a successor pipeline that has `denoising_start` set to 0.8 so that it only denoises the + final 20% of the scheduler. The denoising_end parameter should ideally be utilized when this pipeline + forms a part of a "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image + Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output). + guidance_scale (`float`, *optional*, defaults to 7.5): + Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598). + `guidance_scale` is defined as `w` of equation 2. of [Imagen + Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale > + 1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, + usually at the expense of lower image quality. + negative_prompt (`str` or `List[str]`, *optional*): + The prompt or prompts not to guide the image generation. If not defined, one has to pass + `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is + less than `1`). + negative_prompt_2 (`str` or `List[str]`, *optional*): + The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and + `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders + prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + negative_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input + argument. + pooled_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + negative_pooled_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` + input argument. + ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters. + ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*): + Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of + IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should + contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not + provided, embeddings are computed from the `ip_adapter_image` input argument. + num_images_per_prompt (`int`, *optional*, defaults to 1): + The number of images to generate per prompt. + eta (`float`, *optional*, defaults to 0.0): + Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to + [`schedulers.DDIMScheduler`], will be ignored for others. + generator (`torch.Generator`, *optional*): + One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html) + to make generation deterministic. + latents (`torch.Tensor`, *optional*): + Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image + generation. Can be used to tweak the same generation with different prompts. If not provided, a latents + tensor will ge generated by sampling using the supplied random `generator`. + output_type (`str`, *optional*, defaults to `"pil"`): + The output format of the generate image. Choose between + [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`. + return_dict (`bool`, *optional*, defaults to `True`): + Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a + plain tuple. + cross_attention_kwargs (`dict`, *optional*): + A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under + `self.processor` in + [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). + controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0): + The outputs of the controlnet are multiplied by `controlnet_conditioning_scale` before they are added + to the residual in the original unet. If multiple ControlNets are specified in init, you can set the + corresponding scale as a list. + control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0): + The percentage of total steps at which the controlnet starts applying. + control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0): + The percentage of total steps at which the controlnet stops applying. + original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled. + `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as + explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): + `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position + `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting + `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + For most cases, `target_size` should be set to the desired height and width of the generated image. If + not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in + section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + To negatively condition the generation process based on a specific image resolution. Part of SDXL's + micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): + To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's + micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + To negatively condition the generation process based on a target image resolution. It should be as same + as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + aesthetic_score (`float`, *optional*, defaults to 6.0): + Used to simulate an aesthetic score of the generated image by influencing the positive text condition. + Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + negative_aesthetic_score (`float`, *optional*, defaults to 2.5): + Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to + simulate an aesthetic score of the generated image by influencing the negative text condition. + callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*): + A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of + each denoising step during the inference. with the following arguments: `callback_on_step_end(self: + DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a + list of all tensors as specified by `callback_on_step_end_tensor_inputs`. + callback_on_step_end_tensor_inputs (`List`, *optional*): + The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list + will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the + `._callback_tensor_inputs` attribute of your pipeline class. + + Examples: + + Returns: + [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] or `tuple`: + [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] if `return_dict` is True, otherwise a + `tuple. `tuple. When returning a tuple, the first element is a list with the generated images. + """ + + callback = kwargs.pop("callback", None) + callback_steps = kwargs.pop("callback_steps", None) + + if callback is not None: + deprecate( + "callback", + "1.0.0", + "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", + ) + if callback_steps is not None: + deprecate( + "callback_steps", + "1.0.0", + "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", + ) + + if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)): + callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs + + controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet + + # align format for control guidance + if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list): + control_guidance_start = len(control_guidance_end) * [control_guidance_start] + elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list): + control_guidance_end = len(control_guidance_start) * [control_guidance_end] + elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list): + mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1 + control_guidance_start, control_guidance_end = ( + mult * [control_guidance_start], + mult * [control_guidance_end], + ) + + # from IPython import embed; embed() + # 1. Check inputs. Raise error if not correct + self.check_inputs( + prompt, + control_image, + strength, + num_inference_steps, + callback_steps, + negative_prompt, + prompt_embeds, + negative_prompt_embeds, + pooled_prompt_embeds, + negative_pooled_prompt_embeds, + ip_adapter_image, + ip_adapter_image_embeds, + controlnet_conditioning_scale, + control_guidance_start, + control_guidance_end, + callback_on_step_end_tensor_inputs, + ) + + self._guidance_scale = guidance_scale + self._cross_attention_kwargs = cross_attention_kwargs + self._denoising_end = denoising_end + self._denoising_start = denoising_start + + # 2. Define call parameters + if prompt is not None and isinstance(prompt, str): + batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + device = self._execution_device + + if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float): + controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets) + + # 3.1. Encode input prompt + text_encoder_lora_scale = ( + self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None + ) + ( + prompt_embeds, + negative_prompt_embeds, + pooled_prompt_embeds, + negative_pooled_prompt_embeds, + ) = self.encode_prompt( + prompt, + device, + num_images_per_prompt, + self.do_classifier_free_guidance, + negative_prompt, + prompt_embeds=prompt_embeds, + negative_prompt_embeds=negative_prompt_embeds, + pooled_prompt_embeds=pooled_prompt_embeds, + negative_pooled_prompt_embeds=negative_pooled_prompt_embeds, + lora_scale=text_encoder_lora_scale, + ) + + # 3.2 Encode ip_adapter_image + if ip_adapter_image is not None or ip_adapter_image_embeds is not None: + image_embeds = self.prepare_ip_adapter_image_embeds( + ip_adapter_image, + ip_adapter_image_embeds, + device, + batch_size * num_images_per_prompt, + self.do_classifier_free_guidance, + ) + + # 4. Prepare image, mask, and controlnet_conditioning_image + if isinstance(controlnet, ControlNetModel): + control_image = self.prepare_control_image( + image=control_image, + width=width, + height=height, + batch_size=batch_size * num_images_per_prompt, + num_images_per_prompt=num_images_per_prompt, + device=device, + dtype=controlnet.dtype, + do_classifier_free_guidance=self.do_classifier_free_guidance, + guess_mode=guess_mode, + ) + height, width = control_image.shape[-2:] + elif isinstance(controlnet, MultiControlNetModel): + control_images = [] + + for control_image_ in control_image: + control_image_ = self.prepare_control_image( + image=control_image_, + width=width, + height=height, + batch_size=batch_size * num_images_per_prompt, + num_images_per_prompt=num_images_per_prompt, + device=device, + dtype=controlnet.dtype, + do_classifier_free_guidance=self.do_classifier_free_guidance, + guess_mode=guess_mode, + ) + + control_images.append(control_image_) + + control_image = control_images + height, width = control_image[0].shape[-2:] + else: + assert False + + # 5. set timesteps + def denoising_value_valid(dnv): + return isinstance(dnv, float) and 0 < dnv < 1 + + timesteps, num_inference_steps = retrieve_timesteps( + self.scheduler, num_inference_steps, device, timesteps, sigmas + ) + timesteps, num_inference_steps = self.get_timesteps( + num_inference_steps, + strength, + device, + denoising_start=self.denoising_start if denoising_value_valid(self.denoising_start) else None, + ) + # check that number of inference steps is not < 1 - as this doesn't make sense + if num_inference_steps < 1: + raise ValueError( + f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline" + f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline." + ) + # at which timestep to set the initial noise (n.b. 50% if strength is 0.5) + latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt) + # create a boolean to check if the strength is set to 1. if so then initialise the latents with pure noise + is_strength_max = strength == 1.0 + + # 6. Preprocess mask and image + if padding_mask_crop is not None: + crops_coords = self.mask_processor.get_crop_region(mask_image, width, height, pad=padding_mask_crop) + resize_mode = "fill" + else: + crops_coords = None + resize_mode = "default" + + original_image = image + init_image = self.image_processor.preprocess( + image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode + ) + init_image = init_image.to(dtype=torch.float32) + + mask = self.mask_processor.preprocess( + mask_image, height=height, width=width, resize_mode=resize_mode, crops_coords=crops_coords + ) + + if masked_image_latents is not None: + masked_image = masked_image_latents + elif init_image.shape[1] == 4: + # if images are in latent space, we can't mask it + masked_image = None + else: + masked_image = init_image * (mask < 0.5) + + # 7. Prepare latent variables + num_channels_latents = self.vae.config.latent_channels + num_channels_unet = self.unet.config.in_channels + return_image_latents = num_channels_unet == 4 + + if latents is None: + if strength >= 1.0: + latents = self.prepare_latents_t2i( + batch_size * num_images_per_prompt, + num_channels_latents, + height, + width, + prompt_embeds.dtype, + device, + generator, + latents, + ) + else: + latents = self.prepare_latents( + init_image, + latent_timestep, + batch_size, + num_images_per_prompt, + prompt_embeds.dtype, + device, + generator, + True, + ) + + # 8. Prepare mask latent variables + mask, masked_image_latents = self.prepare_mask_latents( + mask, + masked_image, + batch_size * num_images_per_prompt, + height, + width, + prompt_embeds.dtype, + device, + generator, + self.do_classifier_free_guidance, + ) + + # 9. Check that sizes of mask, masked image and latents match + if num_channels_unet == 9: + # default case for runwayml/stable-diffusion-inpainting + num_channels_mask = mask.shape[1] + num_channels_masked_image = masked_image_latents.shape[1] + if num_channels_latents + num_channels_mask + num_channels_masked_image != self.unet.config.in_channels: + raise ValueError( + f"Incorrect configuration settings! The config of `pipeline.unet`: {self.unet.config} expects" + f" {self.unet.config.in_channels} but received `num_channels_latents`: {num_channels_latents} +" + f" `num_channels_mask`: {num_channels_mask} + `num_channels_masked_image`: {num_channels_masked_image}" + f" = {num_channels_latents + num_channels_masked_image + num_channels_mask}. Please verify the config of" + " `pipeline.unet` or your `mask_image` or `image` input." + ) + elif num_channels_unet != 4: + raise ValueError( + f"The unet {self.unet.__class__} should have either 4 or 9 input channels, not {self.unet.config.in_channels}." + ) + + # 8.1. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline + extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta) + + # 8.2 Create tensor stating which controlnets to keep + controlnet_keep = [] + for i in range(len(timesteps)): + keeps = [ + 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e) + for s, e in zip(control_guidance_start, control_guidance_end) + ] + controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps) + + # 9 Prepare added time ids & embeddings + if isinstance(control_image, list): + original_size = original_size or control_image[0].shape[-2:] + else: + original_size = original_size or control_image.shape[-2:] + target_size = target_size or (height, width) + + add_text_embeds = pooled_prompt_embeds + add_time_ids = self._get_add_time_ids( + original_size, crops_coords_top_left, target_size, dtype=prompt_embeds.dtype + ) + + if self.do_classifier_free_guidance: + prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) + add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0) + add_time_ids = torch.cat([add_time_ids, add_time_ids], dim=0) + + prompt_embeds = prompt_embeds.to(device) + add_text_embeds = add_text_embeds.to(device) + add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1) + + # 10. Denoising loop + num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0) + + if ( + self.denoising_end is not None + and self.denoising_start is not None + and denoising_value_valid(self.denoising_end) + and denoising_value_valid(self.denoising_start) + and self.denoising_start >= self.denoising_end + ): + raise ValueError( + f"`denoising_start`: {self.denoising_start} cannot be larger than or equal to `denoising_end`: " + + f" {self.denoising_end} when using type float." + ) + elif self.denoising_end is not None and denoising_value_valid(self.denoising_end): + discrete_timestep_cutoff = int( + round( + self.scheduler.config.num_train_timesteps + - (self.denoising_end * self.scheduler.config.num_train_timesteps) + ) + ) + num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps))) + timesteps = timesteps[:num_inference_steps] + + # 11.1 Optionally get Guidance Scale Embedding + timestep_cond = None + if self.unet.config.time_cond_proj_dim is not None: + guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt) + timestep_cond = self.get_guidance_scale_embedding( + guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim + ).to(device=device, dtype=latents.dtype) + + # patch diffusers controlnet instance forward, undo + # after denoising loop + + patched_cn_models = [] + if isinstance(self.controlnet, MultiControlNetModel): + cn_models_to_patch = self.controlnet.nets + else: + cn_models_to_patch = [self.controlnet] + + for cn_model in cn_models_to_patch: + cn_og_forward = cn_model.forward + + def _cn_patch_forward(*args, **kwargs): + encoder_hidden_states = kwargs['encoder_hidden_states'] + if cn_model.encoder_hid_proj is not None and cn_model.config.encoder_hid_dim_type == "text_proj": + encoder_hidden_states = cn_model.encoder_hid_proj(kwargs['encoder_hidden_states']) + kwargs.pop('encoder_hidden_states') + return cn_og_forward(*args, encoder_hidden_states=encoder_hidden_states, **kwargs) + + cn_model.forward = _cn_patch_forward + patched_cn_models.append((cn_model, cn_og_forward)) + + try: + with self.progress_bar(total=num_inference_steps) as progress_bar: + for i, t in enumerate(timesteps): + # expand the latents if we are doing classifier free guidance + latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents + latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) + + if num_channels_unet == 9: + latent_model_input = torch.cat([latent_model_input, mask, masked_image_latents], dim=1) + + added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids} + + # controlnet(s) inference + if guess_mode and self.do_classifier_free_guidance: + # Infer ControlNet only for the conditional batch. + control_model_input = latents + control_model_input = self.scheduler.scale_model_input(control_model_input, t) + controlnet_prompt_embeds = prompt_embeds.chunk(2)[1] + controlnet_added_cond_kwargs = { + "text_embeds": add_text_embeds.chunk(2)[1], + "time_ids": add_time_ids.chunk(2)[1], + } + else: + control_model_input = latent_model_input + controlnet_prompt_embeds = prompt_embeds + controlnet_added_cond_kwargs = added_cond_kwargs + + if isinstance(controlnet_keep[i], list): + cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])] + else: + controlnet_cond_scale = controlnet_conditioning_scale + if isinstance(controlnet_cond_scale, list): + controlnet_cond_scale = controlnet_cond_scale[0] + cond_scale = controlnet_cond_scale * controlnet_keep[i] + + down_block_res_samples, mid_block_res_sample = self.controlnet( + control_model_input, + t, + encoder_hidden_states=controlnet_prompt_embeds, + controlnet_cond=control_image, + conditioning_scale=cond_scale, + guess_mode=guess_mode, + added_cond_kwargs=controlnet_added_cond_kwargs, + return_dict=False, + ) + + if guess_mode and self.do_classifier_free_guidance: + # Infered ControlNet only for the conditional batch. + # To apply the output of ControlNet to both the unconditional and conditional batches, + # add 0 to the unconditional batch to keep it unchanged. + down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples] + mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample]) + + if ip_adapter_image is not None or ip_adapter_image_embeds is not None: + added_cond_kwargs["image_embeds"] = image_embeds + + # predict the noise residual + noise_pred = self.unet( + latent_model_input, + t, + encoder_hidden_states=prompt_embeds, + cross_attention_kwargs=self.cross_attention_kwargs, + down_block_additional_residuals=down_block_res_samples, + mid_block_additional_residual=mid_block_res_sample, + added_cond_kwargs=added_cond_kwargs, + return_dict=False, + )[0] + + # perform guidance + if self.do_classifier_free_guidance: + noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) + noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) + + # compute the previous noisy sample x_t -> x_t-1 + latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] + + if callback_on_step_end is not None: + callback_kwargs = {} + for k in callback_on_step_end_tensor_inputs: + callback_kwargs[k] = locals()[k] + callback_outputs = callback_on_step_end(self, i, t, callback_kwargs) + + latents = callback_outputs.pop("latents", latents) + prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) + negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds) + add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds) + negative_pooled_prompt_embeds = callback_outputs.pop( + "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds + ) + add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids) + add_neg_time_ids = callback_outputs.pop("add_neg_time_ids", add_neg_time_ids) + mask = callback_outputs.pop("mask", mask) + masked_image_latents = callback_outputs.pop("masked_image_latents", masked_image_latents) + + # call the callback, if provided + if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): + progress_bar.update() + if callback is not None and i % callback_steps == 0: + step_idx = i // getattr(self.scheduler, "order", 1) + callback(step_idx, t, latents) + finally: + for cn_and_og in patched_cn_models: + cn_and_og[0].forward = cn_and_og[1] + + # If we do sequential model offloading, let's offload unet and controlnet + # manually for max memory savings + if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None: + self.unet.to("cpu") + self.controlnet.to("cpu") + torch.cuda.empty_cache() + torch.cuda.ipc_collect() + + if not output_type == "latent": + # make sure the VAE is in float32 mode, as it overflows in float16 + needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast + + if needs_upcasting: + self.upcast_vae() + latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype) + + latents = latents / self.vae.config.scaling_factor + image = self.vae.decode(latents, return_dict=False)[0] + + # cast back to fp16 if needed + if needs_upcasting: + self.vae.to(dtype=torch.float16) + else: + image = latents + return StableDiffusionXLPipelineOutput(images=image) + + image = self.image_processor.postprocess(image, output_type=output_type) + + # Offload all models + self.maybe_free_model_hooks() + + if not return_dict: + return (image,) + + return StableDiffusionXLPipelineOutput(images=image) diff --git a/examples/community/pipeline_kolors_inpainting.py b/examples/community/pipeline_kolors_inpainting.py new file mode 100644 index 000000000000..aba8535da25d --- /dev/null +++ b/examples/community/pipeline_kolors_inpainting.py @@ -0,0 +1,1726 @@ +# Copyright 2024 The HuggingFace Team. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import inspect +from typing import Any, Callable, Dict, List, Optional, Tuple, Union + +import PIL.Image +import numpy as np +import torch +from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback +from diffusers.image_processor import PipelineImageInput, VaeImageProcessor +from diffusers.loaders import ( + FromSingleFileMixin, + IPAdapterMixin, + StableDiffusionXLLoraLoaderMixin, + TextualInversionLoaderMixin, +) +from diffusers.models import AutoencoderKL, ImageProjection, UNet2DConditionModel +from diffusers.models.attention_processor import ( + AttnProcessor2_0, + LoRAAttnProcessor2_0, + LoRAXFormersAttnProcessor, + XFormersAttnProcessor, +) +from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin +from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput +from diffusers.schedulers import KarrasDiffusionSchedulers +from diffusers.utils import ( + is_invisible_watermark_available, + replace_example_docstring, + is_torch_xla_available, + deprecate, + logging, +) +from diffusers.utils.torch_utils import randn_tensor +from transformers import ( + CLIPImageProcessor, + CLIPVisionModelWithProjection, +) + +from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer + +if is_invisible_watermark_available(): + from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker + +if is_torch_xla_available(): + import torch_xla.core.xla_model as xm + + XLA_AVAILABLE = True +else: + XLA_AVAILABLE = False + +logger = logging.get_logger(__name__) # pylint: disable=invalid-name + + +EXAMPLE_DOC_STRING = """ + Examples: + ```py + >>> import torch + >>> from diffusers import KolorsInpaintPipeline + >>> from diffusers.utils import load_image + + >>> pipe = KolorsInpaintPipeline.from_pretrained( + ... "Kwai-Kolors/Kolors-diffusers", + ... torch_dtype=torch.float16, + ... variant="fp16" + ... ) + >>> pipe.to("cuda") + + >>> img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png" + >>> mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png" + + >>> init_image = load_image(img_url).convert("RGB") + >>> mask_image = load_image(mask_url).convert("RGB") + + >>> prompt = "A majestic tiger sitting on a bench" + >>> image = pipe( + ... prompt=prompt, image=init_image, mask_image=mask_image, num_inference_steps=50, strength=0.80 + ... ).images[0] + ``` +""" + + +# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg +def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0): + """ + Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and + Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4 + """ + std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True) + std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True) + # rescale the results from guidance (fixes overexposure) + noise_pred_rescaled = noise_cfg * (std_text / std_cfg) + # mix with the original results from guidance by factor guidance_rescale to avoid "plain looking" images + noise_cfg = guidance_rescale * noise_pred_rescaled + (1 - guidance_rescale) * noise_cfg + return noise_cfg + + +def mask_pil_to_torch(mask, height, width): + # preprocess mask + if isinstance(mask, (PIL.Image.Image, np.ndarray)): + mask = [mask] + + if isinstance(mask, list) and isinstance(mask[0], PIL.Image.Image): + mask = [i.resize((width, height), resample=PIL.Image.LANCZOS) for i in mask] + mask = np.concatenate([np.array(m.convert("L"))[None, None, :] for m in mask], axis=0) + mask = mask.astype(np.float32) / 255.0 + elif isinstance(mask, list) and isinstance(mask[0], np.ndarray): + mask = np.concatenate([m[None, None, :] for m in mask], axis=0) + + mask = torch.from_numpy(mask) + return mask + + +def prepare_mask_and_masked_image(image, mask, height, width, return_image: bool = False): + """ + Prepares a pair (image, mask) to be consumed by the Stable Diffusion pipeline. This means that those inputs will be + converted to ``torch.Tensor`` with shapes ``batch x channels x height x width`` where ``channels`` is ``3`` for the + ``image`` and ``1`` for the ``mask``. + + The ``image`` will be converted to ``torch.float32`` and normalized to be in ``[-1, 1]``. The ``mask`` will be + binarized (``mask > 0.5``) and cast to ``torch.float32`` too. + + Args: + image (Union[np.array, PIL.Image, torch.Tensor]): The image to inpaint. + It can be a ``PIL.Image``, or a ``height x width x 3`` ``np.array`` or a ``channels x height x width`` + ``torch.Tensor`` or a ``batch x channels x height x width`` ``torch.Tensor``. + mask (_type_): The mask to apply to the image, i.e. regions to inpaint. + It can be a ``PIL.Image``, or a ``height x width`` ``np.array`` or a ``1 x height x width`` + ``torch.Tensor`` or a ``batch x 1 x height x width`` ``torch.Tensor``. + + + Raises: + ValueError: ``torch.Tensor`` images should be in the ``[-1, 1]`` range. ValueError: ``torch.Tensor`` mask + should be in the ``[0, 1]`` range. ValueError: ``mask`` and ``image`` should have the same spatial dimensions. + TypeError: ``mask`` is a ``torch.Tensor`` but ``image`` is not + (ot the other way around). + + Returns: + tuple[torch.Tensor]: The pair (mask, masked_image) as ``torch.Tensor`` with 4 + dimensions: ``batch x channels x height x width``. + """ + + # checkpoint. TOD(Yiyi) - need to clean this up later + deprecation_message = "The prepare_mask_and_masked_image method is deprecated and will be removed in a future version. Please use VaeImageProcessor.preprocess instead" + deprecate( + "prepare_mask_and_masked_image", + "0.30.0", + deprecation_message, + ) + if image is None: + raise ValueError("`image` input cannot be undefined.") + + if mask is None: + raise ValueError("`mask_image` input cannot be undefined.") + + if isinstance(image, torch.Tensor): + if not isinstance(mask, torch.Tensor): + mask = mask_pil_to_torch(mask, height, width) + + if image.ndim == 3: + image = image.unsqueeze(0) + + # Batch and add channel dim for single mask + if mask.ndim == 2: + mask = mask.unsqueeze(0).unsqueeze(0) + + # Batch single mask or add channel dim + if mask.ndim == 3: + # Single batched mask, no channel dim or single mask not batched but channel dim + if mask.shape[0] == 1: + mask = mask.unsqueeze(0) + + # Batched masks no channel dim + else: + mask = mask.unsqueeze(1) + + assert image.ndim == 4 and mask.ndim == 4, "Image and Mask must have 4 dimensions" + # assert image.shape[-2:] == mask.shape[-2:], "Image and Mask must have the same spatial dimensions" + assert image.shape[0] == mask.shape[0], "Image and Mask must have the same batch size" + + # Check image is in [-1, 1] + # if image.min() < -1 or image.max() > 1: + # raise ValueError("Image should be in [-1, 1] range") + + # Check mask is in [0, 1] + if mask.min() < 0 or mask.max() > 1: + raise ValueError("Mask should be in [0, 1] range") + + # Binarize mask + mask[mask < 0.5] = 0 + mask[mask >= 0.5] = 1 + + # Image as float32 + image = image.to(dtype=torch.float32) + elif isinstance(mask, torch.Tensor): + raise TypeError(f"`mask` is a torch.Tensor but `image` (type: {type(image)} is not") + else: + # preprocess image + if isinstance(image, (PIL.Image.Image, np.ndarray)): + image = [image] + if isinstance(image, list) and isinstance(image[0], PIL.Image.Image): + # resize all images w.r.t passed height an width + image = [i.resize((width, height), resample=PIL.Image.LANCZOS) for i in image] + image = [np.array(i.convert("RGB"))[None, :] for i in image] + image = np.concatenate(image, axis=0) + elif isinstance(image, list) and isinstance(image[0], np.ndarray): + image = np.concatenate([i[None, :] for i in image], axis=0) + + image = image.transpose(0, 3, 1, 2) + image = torch.from_numpy(image).to(dtype=torch.float32) / 127.5 - 1.0 + + mask = mask_pil_to_torch(mask, height, width) + mask[mask < 0.5] = 0 + mask[mask >= 0.5] = 1 + + if image.shape[1] == 4: + # images are in latent space and thus can't + # be masked set masked_image to None + # we assume that the checkpoint is not an inpainting + # checkpoint. TOD(Yiyi) - need to clean this up later + masked_image = None + else: + masked_image = image * (mask < 0.5) + + # n.b. ensure backwards compatibility as old function does not return image + if return_image: + return mask, masked_image, image + + return mask, masked_image + + +# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents +def retrieve_latents( + encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" +): + if hasattr(encoder_output, "latent_dist") and sample_mode == "sample": + return encoder_output.latent_dist.sample(generator) + elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax": + return encoder_output.latent_dist.mode() + elif hasattr(encoder_output, "latents"): + return encoder_output.latents + else: + raise AttributeError("Could not access latents of provided encoder_output") + + +# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps +def retrieve_timesteps( + scheduler, + num_inference_steps: Optional[int] = None, + device: Optional[Union[str, torch.device]] = None, + timesteps: Optional[List[int]] = None, + sigmas: Optional[List[float]] = None, + **kwargs, +): + """ + Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles + custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. + + Args: + scheduler (`SchedulerMixin`): + The scheduler to get timesteps from. + num_inference_steps (`int`): + The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps` + must be `None`. + device (`str` or `torch.device`, *optional*): + The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. + timesteps (`List[int]`, *optional*): + Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed, + `num_inference_steps` and `sigmas` must be `None`. + sigmas (`List[float]`, *optional*): + Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed, + `num_inference_steps` and `timesteps` must be `None`. + + Returns: + `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the + second element is the number of inference steps. + """ + if timesteps is not None and sigmas is not None: + raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values") + if timesteps is not None: + accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) + if not accepts_timesteps: + raise ValueError( + f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" + f" timestep schedules. Please check whether you are using the correct scheduler." + ) + scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs) + timesteps = scheduler.timesteps + num_inference_steps = len(timesteps) + elif sigmas is not None: + accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) + if not accept_sigmas: + raise ValueError( + f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" + f" sigmas schedules. Please check whether you are using the correct scheduler." + ) + scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs) + timesteps = scheduler.timesteps + num_inference_steps = len(timesteps) + else: + scheduler.set_timesteps(num_inference_steps, device=device, **kwargs) + timesteps = scheduler.timesteps + return timesteps, num_inference_steps + + +class KolorsInpaintPipeline( + DiffusionPipeline, + StableDiffusionMixin, + StableDiffusionXLLoraLoaderMixin, + FromSingleFileMixin, + IPAdapterMixin, +): + r""" + Pipeline for text-to-image generation using Kolors. + + This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the + library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.) + + The pipeline also inherits the following loading methods: + - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.safetensors` files + - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights + - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights + - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters + + Args: + vae ([`AutoencoderKL`]): + Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. + text_encoder ([`ChatGLMModel`]): + Frozen text-encoder. Kolors uses [ChatGLM3-6B](https://huggingface.co/THUDM/chatglm3-6b). + tokenizer (`ChatGLMTokenizer`): + Tokenizer of class + [ChatGLMTokenizer](https://huggingface.co/THUDM/chatglm3-6b/blob/main/tokenization_chatglm.py). + unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents. + scheduler ([`SchedulerMixin`]): + A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of + [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`]. + requires_aesthetics_score (`bool`, *optional*, defaults to `"False"`): + Whether the `unet` requires a aesthetic_score condition to be passed during inference. + force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`): + Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of + `Kwai-Kolors/Kolors-diffusers`. + add_watermarker (`bool`, *optional*): + Whether to use the [invisible_watermark library](https://github.com/ShieldMnt/invisible-watermark/) to + watermark output images. If not defined, it will default to True if the package is installed, otherwise no + watermarker will be used. + """ + + model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae" + + _optional_components = [ + "tokenizer", + "text_encoder", + "image_encoder", + "feature_extractor", + ] + _callback_tensor_inputs = [ + "latents", + "prompt_embeds", + "negative_prompt_embeds", + "add_text_embeds", + "add_time_ids", + "negative_pooled_prompt_embeds", + "add_neg_time_ids", + "mask", + "masked_image_latents", + ] + + def __init__( + self, + vae: AutoencoderKL, + text_encoder: ChatGLMModel, + tokenizer: ChatGLMTokenizer, + unet: UNet2DConditionModel, + scheduler: KarrasDiffusionSchedulers, + image_encoder: CLIPVisionModelWithProjection = None, + feature_extractor: CLIPImageProcessor = None, + requires_aesthetics_score: bool = False, + force_zeros_for_empty_prompt: bool = True, + add_watermarker: Optional[bool] = None, + ): + super().__init__() + + self.register_modules( + vae=vae, + text_encoder=text_encoder, + tokenizer=tokenizer, + unet=unet, + image_encoder=image_encoder, + feature_extractor=feature_extractor, + scheduler=scheduler, + ) + self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt) + self.register_to_config(requires_aesthetics_score=requires_aesthetics_score) + self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) + self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor) + self.mask_processor = VaeImageProcessor( + vae_scale_factor=self.vae_scale_factor, do_normalize=False, do_binarize=True, do_convert_grayscale=True + ) + + add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available() + + if add_watermarker: + self.watermark = StableDiffusionXLWatermarker() + else: + self.watermark = None + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image + def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None): + dtype = next(self.image_encoder.parameters()).dtype + + if not isinstance(image, torch.Tensor): + image = self.feature_extractor(image, return_tensors="pt").pixel_values + + image = image.to(device=device, dtype=dtype) + if output_hidden_states: + image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2] + image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_enc_hidden_states = self.image_encoder( + torch.zeros_like(image), output_hidden_states=True + ).hidden_states[-2] + uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave( + num_images_per_prompt, dim=0 + ) + return image_enc_hidden_states, uncond_image_enc_hidden_states + else: + image_embeds = self.image_encoder(image).image_embeds + image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) + uncond_image_embeds = torch.zeros_like(image_embeds) + + return image_embeds, uncond_image_embeds + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds + def prepare_ip_adapter_image_embeds( + self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance + ): + if ip_adapter_image_embeds is None: + if not isinstance(ip_adapter_image, list): + ip_adapter_image = [ip_adapter_image] + + if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers): + raise ValueError( + f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters." + ) + + image_embeds = [] + for single_ip_adapter_image, image_proj_layer in zip( + ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers + ): + output_hidden_state = not isinstance(image_proj_layer, ImageProjection) + single_image_embeds, single_negative_image_embeds = self.encode_image( + single_ip_adapter_image, device, 1, output_hidden_state + ) + single_image_embeds = torch.stack([single_image_embeds] * num_images_per_prompt, dim=0) + single_negative_image_embeds = torch.stack( + [single_negative_image_embeds] * num_images_per_prompt, dim=0 + ) + + if do_classifier_free_guidance: + single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds]) + single_image_embeds = single_image_embeds.to(device) + + image_embeds.append(single_image_embeds) + else: + repeat_dims = [1] + image_embeds = [] + for single_image_embeds in ip_adapter_image_embeds: + if do_classifier_free_guidance: + single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2) + single_image_embeds = single_image_embeds.repeat( + num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:])) + ) + single_negative_image_embeds = single_negative_image_embeds.repeat( + num_images_per_prompt, *(repeat_dims * len(single_negative_image_embeds.shape[1:])) + ) + single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds]) + else: + single_image_embeds = single_image_embeds.repeat( + num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:])) + ) + image_embeds.append(single_image_embeds) + + return image_embeds + + def encode_prompt( + self, + prompt, + device: Optional[torch.device] = None, + num_images_per_prompt: int = 1, + do_classifier_free_guidance: bool = True, + negative_prompt=None, + prompt_embeds: Optional[torch.FloatTensor] = None, + negative_prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + lora_scale: Optional[float] = None, + ): + r""" + Encodes the prompt into text encoder hidden states. + + Args: + prompt (`str` or `List[str]`, *optional*): + prompt to be encoded + device: (`torch.device`): + torch device + num_images_per_prompt (`int`): + number of images that should be generated per prompt + do_classifier_free_guidance (`bool`): + whether to use classifier free guidance or not + negative_prompt (`str` or `List[str]`, *optional*): + The prompt or prompts not to guide the image generation. If not defined, one has to pass + `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is + less than `1`). + prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + negative_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input + argument. + pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` + input argument. + lora_scale (`float`, *optional*): + A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. + """ + device = device or self._execution_device + + # set lora scale so that monkey patched LoRA + # function of text encoder can correctly access it + if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin): + self._lora_scale = lora_scale + + if prompt is not None and isinstance(prompt, str): + batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + # Define tokenizers and text encoders + tokenizers = [self.tokenizer] + text_encoders = [self.text_encoder] + + if prompt_embeds is None: + # textual inversion: procecss multi-vector tokens if necessary + prompt_embeds_list = [] + for tokenizer, text_encoder in zip(tokenizers, text_encoders): + if isinstance(self, TextualInversionLoaderMixin): + prompt = self.maybe_convert_prompt(prompt, tokenizer) + + text_inputs = tokenizer( + prompt, + padding="max_length", + max_length=256, + truncation=True, + return_tensors="pt", + ).to(self._execution_device) + output = text_encoder( + input_ids=text_inputs['input_ids'], + attention_mask=text_inputs['attention_mask'], + position_ids=text_inputs['position_ids'], + output_hidden_states=True) + prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() + pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] + bs_embed, seq_len, _ = prompt_embeds.shape + prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) + prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1) + prompt_embeds_list.append(prompt_embeds) + + # prompt_embeds = torch.concat(prompt_embeds_list, dim=-1) + prompt_embeds = prompt_embeds_list[0] + + # get unconditional embeddings for classifier free guidance + zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt + if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt: + negative_prompt_embeds = torch.zeros_like(prompt_embeds) + negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds) + elif do_classifier_free_guidance and negative_prompt_embeds is None: + # negative_prompt = negative_prompt or "" + uncond_tokens: List[str] + if negative_prompt is None: + uncond_tokens = [""] * batch_size + elif prompt is not None and type(prompt) is not type(negative_prompt): + raise TypeError( + f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" + f" {type(prompt)}." + ) + elif isinstance(negative_prompt, str): + uncond_tokens = [negative_prompt] + elif batch_size != len(negative_prompt): + raise ValueError( + f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" + f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" + " the batch size of `prompt`." + ) + else: + uncond_tokens = negative_prompt + + negative_prompt_embeds_list = [] + for tokenizer, text_encoder in zip(tokenizers, text_encoders): + # textual inversion: procecss multi-vector tokens if necessary + if isinstance(self, TextualInversionLoaderMixin): + uncond_tokens = self.maybe_convert_prompt(uncond_tokens, tokenizer) + + max_length = prompt_embeds.shape[1] + uncond_input = tokenizer( + uncond_tokens, + padding="max_length", + max_length=max_length, + truncation=True, + return_tensors="pt", + ).to(self._execution_device) + output = text_encoder( + input_ids=uncond_input['input_ids'], + attention_mask=uncond_input['attention_mask'], + position_ids=uncond_input['position_ids'], + output_hidden_states=True) + negative_prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() + negative_pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] + + if do_classifier_free_guidance: + # duplicate unconditional embeddings for each generation per prompt, using mps friendly method + seq_len = negative_prompt_embeds.shape[1] + + negative_prompt_embeds = negative_prompt_embeds.to(dtype=text_encoder.dtype, device=device) + + negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1) + negative_prompt_embeds = negative_prompt_embeds.view( + batch_size * num_images_per_prompt, seq_len, -1 + ) + + # For classifier free guidance, we need to do two forward passes. + # Here we concatenate the unconditional and text embeddings into a single batch + # to avoid doing two forward passes + + negative_prompt_embeds_list.append(negative_prompt_embeds) + + # negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1) + negative_prompt_embeds = negative_prompt_embeds_list[0] + + bs_embed = pooled_prompt_embeds.shape[0] + pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( + bs_embed * num_images_per_prompt, -1 + ) + if do_classifier_free_guidance: + negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( + bs_embed * num_images_per_prompt, -1 + ) + + return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs + def prepare_extra_step_kwargs(self, generator, eta): + # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature + # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. + # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 + # and should be between [0, 1] + + accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) + extra_step_kwargs = {} + if accepts_eta: + extra_step_kwargs["eta"] = eta + + # check if the scheduler accepts generator + accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) + if accepts_generator: + extra_step_kwargs["generator"] = generator + return extra_step_kwargs + + def check_inputs( + self, + prompt, + image, + mask_image, + height, + width, + strength, + callback_steps, + output_type, + negative_prompt=None, + prompt_embeds=None, + negative_prompt_embeds=None, + ip_adapter_image=None, + ip_adapter_image_embeds=None, + callback_on_step_end_tensor_inputs=None, + padding_mask_crop=None, + ): + if strength < 0 or strength > 1: + raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}") + + if height % 8 != 0 or width % 8 != 0: + raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.") + + if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0): + raise ValueError( + f"`callback_steps` has to be a positive integer but is {callback_steps} of type" + f" {type(callback_steps)}." + ) + + if callback_on_step_end_tensor_inputs is not None and not all( + k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs + ): + raise ValueError( + f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" + ) + + if prompt is not None and prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to" + " only forward one of the two." + ) + elif prompt is None and prompt_embeds is None: + raise ValueError( + "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." + ) + elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): + raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") + + if negative_prompt is not None and negative_prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:" + f" {negative_prompt_embeds}. Please make sure to only forward one of the two." + ) + + if prompt_embeds is not None and negative_prompt_embeds is not None: + if prompt_embeds.shape != negative_prompt_embeds.shape: + raise ValueError( + "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but" + f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`" + f" {negative_prompt_embeds.shape}." + ) + if padding_mask_crop is not None: + if not isinstance(image, PIL.Image.Image): + raise ValueError( + f"The image should be a PIL image when inpainting mask crop, but is of type" f" {type(image)}." + ) + if not isinstance(mask_image, PIL.Image.Image): + raise ValueError( + f"The mask image should be a PIL image when inpainting mask crop, but is of type" + f" {type(mask_image)}." + ) + if output_type != "pil": + raise ValueError(f"The output type should be PIL when inpainting mask crop, but is" f" {output_type}.") + + if ip_adapter_image is not None and ip_adapter_image_embeds is not None: + raise ValueError( + "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined." + ) + + if ip_adapter_image_embeds is not None: + if not isinstance(ip_adapter_image_embeds, list): + raise ValueError( + f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}" + ) + elif ip_adapter_image_embeds[0].ndim not in [3, 4]: + raise ValueError( + f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D" + ) + + def prepare_latents( + self, + batch_size, + num_channels_latents, + height, + width, + dtype, + device, + generator, + latents=None, + image=None, + timestep=None, + is_strength_max=True, + add_noise=True, + return_noise=False, + return_image_latents=False, + ): + shape = ( + batch_size, + num_channels_latents, + int(height) // self.vae_scale_factor, + int(width) // self.vae_scale_factor, + ) + if isinstance(generator, list) and len(generator) != batch_size: + raise ValueError( + f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" + f" size of {batch_size}. Make sure the batch size matches the length of the generators." + ) + + if (image is None or timestep is None) and not is_strength_max: + raise ValueError( + "Since strength < 1. initial latents are to be initialised as a combination of Image + Noise." + "However, either the image or the noise timestep has not been provided." + ) + + if image.shape[1] == 4: + image_latents = image.to(device=device, dtype=dtype) + image_latents = image_latents.repeat(batch_size // image_latents.shape[0], 1, 1, 1) + elif return_image_latents or (latents is None and not is_strength_max): + image = image.to(device=device, dtype=dtype) + image_latents = self._encode_vae_image(image=image, generator=generator) + image_latents = image_latents.repeat(batch_size // image_latents.shape[0], 1, 1, 1) + + if latents is None and add_noise: + noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype) + # if strength is 1. then initialise the latents to noise, else initial to image + noise + latents = noise if is_strength_max else self.scheduler.add_noise(image_latents, noise, timestep) + # if pure noise then scale the initial latents by the Scheduler's init sigma + latents = latents * self.scheduler.init_noise_sigma if is_strength_max else latents + elif add_noise: + noise = latents.to(device) + latents = noise * self.scheduler.init_noise_sigma + else: + noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype) + latents = image_latents.to(device) + + outputs = (latents,) + + if return_noise: + outputs += (noise,) + + if return_image_latents: + outputs += (image_latents,) + + return outputs + + def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator): + dtype = image.dtype + if self.vae.config.force_upcast: + image = image.float() + self.vae.to(dtype=torch.float32) + + if isinstance(generator, list): + image_latents = [ + retrieve_latents(self.vae.encode(image[i: i + 1]), generator=generator[i]) + for i in range(image.shape[0]) + ] + image_latents = torch.cat(image_latents, dim=0) + else: + image_latents = retrieve_latents(self.vae.encode(image), generator=generator) + + if self.vae.config.force_upcast: + self.vae.to(dtype) + + image_latents = image_latents.to(dtype) + image_latents = self.vae.config.scaling_factor * image_latents + + return image_latents + + def prepare_mask_latents( + self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance + ): + # resize the mask to latents shape as we concatenate the mask to the latents + # we do that before converting to dtype to avoid breaking in case we're using cpu_offload + # and half precision + mask = torch.nn.functional.interpolate( + mask, size=(height // self.vae_scale_factor, width // self.vae_scale_factor) + ) + mask = mask.to(device=device, dtype=dtype) + + # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method + if mask.shape[0] < batch_size: + if not batch_size % mask.shape[0] == 0: + raise ValueError( + "The passed mask and the required batch size don't match. Masks are supposed to be duplicated to" + f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number" + " of masks that you pass is divisible by the total requested batch size." + ) + mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1) + + mask = torch.cat([mask] * 2) if do_classifier_free_guidance else mask + + if masked_image is not None and masked_image.shape[1] == 4: + masked_image_latents = masked_image + else: + masked_image_latents = None + + if masked_image is not None: + if masked_image_latents is None: + masked_image = masked_image.to(device=device, dtype=dtype) + masked_image_latents = self._encode_vae_image(masked_image, generator=generator) + + if masked_image_latents.shape[0] < batch_size: + if not batch_size % masked_image_latents.shape[0] == 0: + raise ValueError( + "The passed images and the required batch size don't match. Images are supposed to be duplicated" + f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed." + " Make sure the number of images that you pass is divisible by the total requested batch size." + ) + masked_image_latents = masked_image_latents.repeat( + batch_size // masked_image_latents.shape[0], 1, 1, 1 + ) + + masked_image_latents = ( + torch.cat([masked_image_latents] * 2) if do_classifier_free_guidance else masked_image_latents + ) + + # aligning device to prevent device errors when concating it with the latent model input + masked_image_latents = masked_image_latents.to(device=device, dtype=dtype) + + return mask, masked_image_latents + + # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.get_timesteps + def get_timesteps(self, num_inference_steps, strength, device, denoising_start=None): + # get the original timestep using init_timestep + if denoising_start is None: + init_timestep = min(int(num_inference_steps * strength), num_inference_steps) + t_start = max(num_inference_steps - init_timestep, 0) + else: + t_start = 0 + + timesteps = self.scheduler.timesteps[t_start * self.scheduler.order:] + + # Strength is irrelevant if we directly request a timestep to start at; + # that is, strength is determined by the denoising_start instead. + if denoising_start is not None: + discrete_timestep_cutoff = int( + round( + self.scheduler.config.num_train_timesteps + - (denoising_start * self.scheduler.config.num_train_timesteps) + ) + ) + + num_inference_steps = (timesteps < discrete_timestep_cutoff).sum().item() + if self.scheduler.order == 2 and num_inference_steps % 2 == 0: + # if the scheduler is a 2nd order scheduler we might have to do +1 + # because `num_inference_steps` might be even given that every timestep + # (except the highest one) is duplicated. If `num_inference_steps` is even it would + # mean that we cut the timesteps in the middle of the denoising step + # (between 1st and 2nd derivative) which leads to incorrect results. By adding 1 + # we ensure that the denoising process always ends after the 2nd derivate step of the scheduler + num_inference_steps = num_inference_steps + 1 + + # because t_n+1 >= t_n, we slice the timesteps starting from the end + timesteps = timesteps[-num_inference_steps:] + return timesteps, num_inference_steps + + return timesteps, num_inference_steps - t_start + + # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline._get_add_time_ids + def _get_add_time_ids( + self, + original_size, + crops_coords_top_left, + target_size, + aesthetic_score, + negative_aesthetic_score, + negative_original_size, + negative_crops_coords_top_left, + negative_target_size, + dtype, + text_encoder_projection_dim=None, + ): + if self.config.requires_aesthetics_score: + add_time_ids = list(original_size + crops_coords_top_left + (aesthetic_score,)) + add_neg_time_ids = list( + negative_original_size + negative_crops_coords_top_left + (negative_aesthetic_score,) + ) + else: + add_time_ids = list(original_size + crops_coords_top_left + target_size) + add_neg_time_ids = list(negative_original_size + crops_coords_top_left + negative_target_size) + + passed_add_embed_dim = ( + self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 + ) + expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features + + if ( + expected_add_embed_dim > passed_add_embed_dim + and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim + ): + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to enable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=True)` to make sure `aesthetic_score` {aesthetic_score} and `negative_aesthetic_score` {negative_aesthetic_score} is correctly used by the model." + ) + elif ( + expected_add_embed_dim < passed_add_embed_dim + and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim + ): + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to disable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=False)` to make sure `target_size` {target_size} is correctly used by the model." + ) + elif expected_add_embed_dim != passed_add_embed_dim: + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder.config.projection_dim`." + ) + + add_time_ids = torch.tensor([add_time_ids], dtype=dtype) + add_neg_time_ids = torch.tensor([add_neg_time_ids], dtype=dtype) + + return add_time_ids, add_neg_time_ids + + # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae + def upcast_vae(self): + dtype = self.vae.dtype + self.vae.to(dtype=torch.float32) + use_torch_2_0_or_xformers = isinstance( + self.vae.decoder.mid_block.attentions[0].processor, + ( + AttnProcessor2_0, + XFormersAttnProcessor, + LoRAXFormersAttnProcessor, + LoRAAttnProcessor2_0, + ), + ) + # if xformers or torch_2_0 is used attention block does not need + # to be in float32 which can save lots of memory + if use_torch_2_0_or_xformers: + self.vae.post_quant_conv.to(dtype) + self.vae.decoder.conv_in.to(dtype) + self.vae.decoder.mid_block.to(dtype) + + # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding + def get_guidance_scale_embedding( + self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32 + ) -> torch.Tensor: + """ + See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298 + + Args: + w (`torch.Tensor`): + Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings. + embedding_dim (`int`, *optional*, defaults to 512): + Dimension of the embeddings to generate. + dtype (`torch.dtype`, *optional*, defaults to `torch.float32`): + Data type of the generated embeddings. + + Returns: + `torch.Tensor`: Embedding vectors with shape `(len(w), embedding_dim)`. + """ + assert len(w.shape) == 1 + w = w * 1000.0 + + half_dim = embedding_dim // 2 + emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1) + emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb) + emb = w.to(dtype)[:, None] * emb[None, :] + emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1) + if embedding_dim % 2 == 1: # zero pad + emb = torch.nn.functional.pad(emb, (0, 1)) + assert emb.shape == (w.shape[0], embedding_dim) + return emb + + @property + def guidance_scale(self): + return self._guidance_scale + + @property + def guidance_rescale(self): + return self._guidance_rescale + + # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2) + # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1` + # corresponds to doing no classifier free guidance. + @property + def do_classifier_free_guidance(self): + return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None + + @property + def cross_attention_kwargs(self): + return self._cross_attention_kwargs + + @property + def denoising_end(self): + return self._denoising_end + + @property + def denoising_start(self): + return self._denoising_start + + @property + def num_timesteps(self): + return self._num_timesteps + + @property + def interrupt(self): + return self._interrupt + + @torch.no_grad() + @replace_example_docstring(EXAMPLE_DOC_STRING) + def __call__( + self, + prompt: Union[str, List[str]] = None, + image: PipelineImageInput = None, + mask_image: PipelineImageInput = None, + masked_image_latents: torch.Tensor = None, + height: Optional[int] = None, + width: Optional[int] = None, + padding_mask_crop: Optional[int] = None, + strength: float = 0.9999, + num_inference_steps: int = 50, + timesteps: List[int] = None, + sigmas: List[float] = None, + denoising_start: Optional[float] = None, + denoising_end: Optional[float] = None, + guidance_scale: float = 7.5, + negative_prompt: Optional[Union[str, List[str]]] = None, + num_images_per_prompt: Optional[int] = 1, + eta: float = 0.0, + generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, + latents: Optional[torch.Tensor] = None, + prompt_embeds: Optional[torch.Tensor] = None, + negative_prompt_embeds: Optional[torch.Tensor] = None, + pooled_prompt_embeds: Optional[torch.Tensor] = None, + negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, + ip_adapter_image: Optional[PipelineImageInput] = None, + ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, + output_type: Optional[str] = "pil", + return_dict: bool = True, + cross_attention_kwargs: Optional[Dict[str, Any]] = None, + guidance_rescale: float = 0.0, + original_size: Tuple[int, int] = None, + crops_coords_top_left: Tuple[int, int] = (0, 0), + target_size: Tuple[int, int] = None, + negative_original_size: Optional[Tuple[int, int]] = None, + negative_crops_coords_top_left: Tuple[int, int] = (0, 0), + negative_target_size: Optional[Tuple[int, int]] = None, + aesthetic_score: float = 6.0, + negative_aesthetic_score: float = 2.5, + callback_on_step_end: Optional[ + Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] + ] = None, + callback_on_step_end_tensor_inputs: List[str] = ["latents"], + **kwargs, + ): + r""" + Function invoked when calling the pipeline for generation. + + Args: + prompt (`str` or `List[str]`, *optional*): + The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`. + instead. + image (`PIL.Image.Image`): + `Image`, or tensor representing an image batch which will be inpainted, *i.e.* parts of the image will + be masked out with `mask_image` and repainted according to `prompt`. + mask_image (`PIL.Image.Image`): + `Image`, or tensor representing an image batch, to mask `image`. White pixels in the mask will be + repainted, while black pixels will be preserved. If `mask_image` is a PIL image, it will be converted + to a single channel (luminance) before use. If it's a tensor, it should contain one color channel (L) + instead of 3, so the expected shape would be `(B, H, W, 1)`. + height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor): + The height in pixels of the generated image. This is set to 1024 by default for the best results. + Anything below 512 pixels won't work well for + [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) + and checkpoints that are not specifically fine-tuned on low resolutions. + width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor): + The width in pixels of the generated image. This is set to 1024 by default for the best results. + Anything below 512 pixels won't work well for + [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) + and checkpoints that are not specifically fine-tuned on low resolutions. + padding_mask_crop (`int`, *optional*, defaults to `None`): + The size of margin in the crop to be applied to the image and masking. If `None`, no crop is applied to + image and mask_image. If `padding_mask_crop` is not `None`, it will first find a rectangular region + with the same aspect ration of the image and contains all masked area, and then expand that area based + on `padding_mask_crop`. The image and mask_image will then be cropped based on the expanded area before + resizing to the original image size for inpainting. This is useful when the masked area is small while + the image is large and contain information irrelevant for inpainting, such as background. + strength (`float`, *optional*, defaults to 0.9999): + Conceptually, indicates how much to transform the masked portion of the reference `image`. Must be + between 0 and 1. `image` will be used as a starting point, adding more noise to it the larger the + `strength`. The number of denoising steps depends on the amount of noise initially added. When + `strength` is 1, added noise will be maximum and the denoising process will run for the full number of + iterations specified in `num_inference_steps`. A value of 1, therefore, essentially ignores the masked + portion of the reference `image`. Note that in the case of `denoising_start` being declared as an + integer, the value of `strength` will be ignored. + num_inference_steps (`int`, *optional*, defaults to 50): + The number of denoising steps. More denoising steps usually lead to a higher quality image at the + expense of slower inference. + timesteps (`List[int]`, *optional*): + Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument + in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is + passed will be used. Must be in descending order. + sigmas (`List[float]`, *optional*): + Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in + their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed + will be used. + denoising_start (`float`, *optional*): + When specified, indicates the fraction (between 0.0 and 1.0) of the total denoising process to be + bypassed before it is initiated. Consequently, the initial part of the denoising process is skipped and + it is assumed that the passed `image` is a partly denoised image. Note that when this is specified, + strength will be ignored. The `denoising_start` parameter is particularly beneficial when this pipeline + is integrated into a "Mixture of Denoisers" multi-pipeline setup, as detailed in [**Refining the Image + Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output). + denoising_end (`float`, *optional*): + When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be + completed before it is intentionally prematurely terminated. As a result, the returned sample will + still retain a substantial amount of noise (ca. final 20% of timesteps still needed) and should be + denoised by a successor pipeline that has `denoising_start` set to 0.8 so that it only denoises the + final 20% of the scheduler. The denoising_end parameter should ideally be utilized when this pipeline + forms a part of a "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image + Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output). + guidance_scale (`float`, *optional*, defaults to 7.5): + Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598). + `guidance_scale` is defined as `w` of equation 2. of [Imagen + Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale > + 1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, + usually at the expense of lower image quality. + negative_prompt (`str` or `List[str]`, *optional*): + The prompt or prompts not to guide the image generation. If not defined, one has to pass + `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is + less than `1`). + prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + negative_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input + argument. + pooled_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + negative_pooled_prompt_embeds (`torch.Tensor`, *optional*): + Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt + weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` + input argument. + ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters. + ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*): + Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of + IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should + contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not + provided, embeddings are computed from the `ip_adapter_image` input argument. + num_images_per_prompt (`int`, *optional*, defaults to 1): + The number of images to generate per prompt. + eta (`float`, *optional*, defaults to 0.0): + Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to + [`schedulers.DDIMScheduler`], will be ignored for others. + generator (`torch.Generator`, *optional*): + One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html) + to make generation deterministic. + latents (`torch.Tensor`, *optional*): + Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image + generation. Can be used to tweak the same generation with different prompts. If not provided, a latents + tensor will ge generated by sampling using the supplied random `generator`. + output_type (`str`, *optional*, defaults to `"pil"`): + The output format of the generate image. Choose between + [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`. + return_dict (`bool`, *optional*, defaults to `True`): + Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a + plain tuple. + cross_attention_kwargs (`dict`, *optional*): + A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under + `self.processor` in + [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). + original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled. + `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as + explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): + `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position + `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting + `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + For most cases, `target_size` should be set to the desired height and width of the generated image. If + not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in + section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + To negatively condition the generation process based on a specific image resolution. Part of SDXL's + micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): + To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's + micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + To negatively condition the generation process based on a target image resolution. It should be as same + as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + aesthetic_score (`float`, *optional*, defaults to 6.0): + Used to simulate an aesthetic score of the generated image by influencing the positive text condition. + Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + negative_aesthetic_score (`float`, *optional*, defaults to 2.5): + Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to + simulate an aesthetic score of the generated image by influencing the negative text condition. + callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*): + A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of + each denoising step during the inference. with the following arguments: `callback_on_step_end(self: + DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a + list of all tensors as specified by `callback_on_step_end_tensor_inputs`. + callback_on_step_end_tensor_inputs (`List`, *optional*): + The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list + will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the + `._callback_tensor_inputs` attribute of your pipeline class. + + Examples: + + Returns: + [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] or `tuple`: + [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] if `return_dict` is True, otherwise a + `tuple. `tuple. When returning a tuple, the first element is a list with the generated images. + """ + + callback = kwargs.pop("callback", None) + callback_steps = kwargs.pop("callback_steps", None) + + if callback is not None: + deprecate( + "callback", + "1.0.0", + "Passing `callback` as an input argument to `__call__` is deprecated, consider use `callback_on_step_end`", + ) + if callback_steps is not None: + deprecate( + "callback_steps", + "1.0.0", + "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider use `callback_on_step_end`", + ) + + if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)): + callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs + + # 0. Default height and width to unet + height = height or self.unet.config.sample_size * self.vae_scale_factor + width = width or self.unet.config.sample_size * self.vae_scale_factor + + # 1. Check inputs + self.check_inputs( + prompt, + image, + mask_image, + height, + width, + strength, + callback_steps, + output_type, + negative_prompt, + prompt_embeds, + negative_prompt_embeds, + ip_adapter_image, + ip_adapter_image_embeds, + callback_on_step_end_tensor_inputs, + padding_mask_crop, + ) + + self._guidance_scale = guidance_scale + self._guidance_rescale = guidance_rescale + self._cross_attention_kwargs = cross_attention_kwargs + self._denoising_end = denoising_end + self._denoising_start = denoising_start + self._interrupt = False + + # 2. Define call parameters + if prompt is not None and isinstance(prompt, str): + batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + device = self._execution_device + + # 3. Encode input prompt + text_encoder_lora_scale = ( + self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None + ) + + ( + prompt_embeds, + negative_prompt_embeds, + pooled_prompt_embeds, + negative_pooled_prompt_embeds, + ) = self.encode_prompt( + prompt=prompt, + device=device, + num_images_per_prompt=num_images_per_prompt, + do_classifier_free_guidance=self.do_classifier_free_guidance, + negative_prompt=negative_prompt, + prompt_embeds=prompt_embeds, + negative_prompt_embeds=negative_prompt_embeds, + pooled_prompt_embeds=pooled_prompt_embeds, + negative_pooled_prompt_embeds=negative_pooled_prompt_embeds, + lora_scale=text_encoder_lora_scale, + ) + + # 4. set timesteps + def denoising_value_valid(dnv): + return isinstance(dnv, float) and 0 < dnv < 1 + + timesteps, num_inference_steps = retrieve_timesteps( + self.scheduler, num_inference_steps, device, timesteps, sigmas + ) + timesteps, num_inference_steps = self.get_timesteps( + num_inference_steps, + strength, + device, + denoising_start=self.denoising_start if denoising_value_valid(self.denoising_start) else None, + ) + # check that number of inference steps is not < 1 - as this doesn't make sense + if num_inference_steps < 1: + raise ValueError( + f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline" + f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline." + ) + # at which timestep to set the initial noise (n.b. 50% if strength is 0.5) + latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt) + # create a boolean to check if the strength is set to 1. if so then initialise the latents with pure noise + is_strength_max = strength == 1.0 + + # 5. Preprocess mask and image + if padding_mask_crop is not None: + crops_coords = self.mask_processor.get_crop_region(mask_image, width, height, pad=padding_mask_crop) + resize_mode = "fill" + else: + crops_coords = None + resize_mode = "default" + + original_image = image + init_image = self.image_processor.preprocess( + image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode + ) + init_image = init_image.to(dtype=torch.float32) + + mask = self.mask_processor.preprocess( + mask_image, height=height, width=width, resize_mode=resize_mode, crops_coords=crops_coords + ) + + if masked_image_latents is not None: + masked_image = masked_image_latents + elif init_image.shape[1] == 4: + # if images are in latent space, we can't mask it + masked_image = None + else: + masked_image = init_image * (mask < 0.5) + + # 6. Prepare latent variables + num_channels_latents = self.vae.config.latent_channels + num_channels_unet = self.unet.config.in_channels + return_image_latents = num_channels_unet == 4 + + add_noise = True if self.denoising_start is None else False + latents_outputs = self.prepare_latents( + batch_size * num_images_per_prompt, + num_channels_latents, + height, + width, + prompt_embeds.dtype, + device, + generator, + latents, + image=init_image, + timestep=latent_timestep, + is_strength_max=is_strength_max, + add_noise=add_noise, + return_noise=True, + return_image_latents=return_image_latents, + ) + + if return_image_latents: + latents, noise, image_latents = latents_outputs + else: + latents, noise = latents_outputs + + # 7. Prepare mask latent variables + mask, masked_image_latents = self.prepare_mask_latents( + mask, + masked_image, + batch_size * num_images_per_prompt, + height, + width, + prompt_embeds.dtype, + device, + generator, + self.do_classifier_free_guidance, + ) + + # 8. Check that sizes of mask, masked image and latents match + if num_channels_unet == 9: + # default case for runwayml/stable-diffusion-inpainting + num_channels_mask = mask.shape[1] + num_channels_masked_image = masked_image_latents.shape[1] + if num_channels_latents + num_channels_mask + num_channels_masked_image != self.unet.config.in_channels: + raise ValueError( + f"Incorrect configuration settings! The config of `pipeline.unet`: {self.unet.config} expects" + f" {self.unet.config.in_channels} but received `num_channels_latents`: {num_channels_latents} +" + f" `num_channels_mask`: {num_channels_mask} + `num_channels_masked_image`: {num_channels_masked_image}" + f" = {num_channels_latents + num_channels_masked_image + num_channels_mask}. Please verify the config of" + " `pipeline.unet` or your `mask_image` or `image` input." + ) + elif num_channels_unet != 4: + raise ValueError( + f"The unet {self.unet.__class__} should have either 4 or 9 input channels, not {self.unet.config.in_channels}." + ) + # 8.1 Prepare extra step kwargs. + extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta) + + # 9. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline + height, width = latents.shape[-2:] + height = height * self.vae_scale_factor + width = width * self.vae_scale_factor + + original_size = original_size or (height, width) + target_size = target_size or (height, width) + + # 10. Prepare added time ids & embeddings + if negative_original_size is None: + negative_original_size = original_size + if negative_target_size is None: + negative_target_size = target_size + + add_text_embeds = pooled_prompt_embeds + text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1]) + + add_time_ids, add_neg_time_ids = self._get_add_time_ids( + original_size, + crops_coords_top_left, + target_size, + aesthetic_score, + negative_aesthetic_score, + negative_original_size, + negative_crops_coords_top_left, + negative_target_size, + dtype=prompt_embeds.dtype, + text_encoder_projection_dim=text_encoder_projection_dim, + ) + add_time_ids = add_time_ids.repeat(batch_size * num_images_per_prompt, 1) + + if self.do_classifier_free_guidance: + prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) + add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0) + add_neg_time_ids = add_neg_time_ids.repeat(batch_size * num_images_per_prompt, 1) + add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0) + + prompt_embeds = prompt_embeds.to(device) + add_text_embeds = add_text_embeds.to(device) + add_time_ids = add_time_ids.to(device) + + if ip_adapter_image is not None or ip_adapter_image_embeds is not None: + image_embeds = self.prepare_ip_adapter_image_embeds( + ip_adapter_image, + ip_adapter_image_embeds, + device, + batch_size * num_images_per_prompt, + self.do_classifier_free_guidance, + ) + + # 11. Denoising loop + num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0) + + if ( + self.denoising_end is not None + and self.denoising_start is not None + and denoising_value_valid(self.denoising_end) + and denoising_value_valid(self.denoising_start) + and self.denoising_start >= self.denoising_end + ): + raise ValueError( + f"`denoising_start`: {self.denoising_start} cannot be larger than or equal to `denoising_end`: " + + f" {self.denoising_end} when using type float." + ) + elif self.denoising_end is not None and denoising_value_valid(self.denoising_end): + discrete_timestep_cutoff = int( + round( + self.scheduler.config.num_train_timesteps + - (self.denoising_end * self.scheduler.config.num_train_timesteps) + ) + ) + num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps))) + timesteps = timesteps[:num_inference_steps] + + # 11.1 Optionally get Guidance Scale Embedding + timestep_cond = None + if self.unet.config.time_cond_proj_dim is not None: + guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt) + timestep_cond = self.get_guidance_scale_embedding( + guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim + ).to(device=device, dtype=latents.dtype) + + self._num_timesteps = len(timesteps) + with self.progress_bar(total=num_inference_steps) as progress_bar: + for i, t in enumerate(timesteps): + if self.interrupt: + continue + # expand the latents if we are doing classifier free guidance + latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents + + # concat latents, mask, masked_image_latents in the channel dimension + latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) + + if num_channels_unet == 9: + latent_model_input = torch.cat([latent_model_input, mask, masked_image_latents], dim=1) + + # predict the noise residual + added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids} + if ip_adapter_image is not None or ip_adapter_image_embeds is not None: + added_cond_kwargs["image_embeds"] = image_embeds + noise_pred = self.unet( + latent_model_input, + t, + encoder_hidden_states=prompt_embeds, + timestep_cond=timestep_cond, + cross_attention_kwargs=self.cross_attention_kwargs, + added_cond_kwargs=added_cond_kwargs, + return_dict=False, + )[0] + + # perform guidance + if self.do_classifier_free_guidance: + noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) + noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond) + + if self.do_classifier_free_guidance and self.guidance_rescale > 0.0: + # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf + noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=self.guidance_rescale) + + # compute the previous noisy sample x_t -> x_t-1 + latents_dtype = latents.dtype + latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] + if latents.dtype != latents_dtype: + if torch.backends.mps.is_available(): + # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272 + latents = latents.to(latents_dtype) + + if num_channels_unet == 4: + init_latents_proper = image_latents + if self.do_classifier_free_guidance: + init_mask, _ = mask.chunk(2) + else: + init_mask = mask + + if i < len(timesteps) - 1: + noise_timestep = timesteps[i + 1] + init_latents_proper = self.scheduler.add_noise( + init_latents_proper, noise, torch.tensor([noise_timestep]) + ) + + latents = (1 - init_mask) * init_latents_proper + init_mask * latents + + if callback_on_step_end is not None: + callback_kwargs = {} + for k in callback_on_step_end_tensor_inputs: + callback_kwargs[k] = locals()[k] + callback_outputs = callback_on_step_end(self, i, t, callback_kwargs) + + latents = callback_outputs.pop("latents", latents) + prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) + negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds) + add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds) + negative_pooled_prompt_embeds = callback_outputs.pop( + "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds + ) + add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids) + add_neg_time_ids = callback_outputs.pop("add_neg_time_ids", add_neg_time_ids) + mask = callback_outputs.pop("mask", mask) + masked_image_latents = callback_outputs.pop("masked_image_latents", masked_image_latents) + + # call the callback, if provided + if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): + progress_bar.update() + if callback is not None and i % callback_steps == 0: + step_idx = i // getattr(self.scheduler, "order", 1) + callback(step_idx, t, latents) + + if XLA_AVAILABLE: + xm.mark_step() + + if not output_type == "latent": + # make sure the VAE is in float32 mode, as it overflows in float16 + needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast + + if needs_upcasting: + self.upcast_vae() + latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype) + elif latents.dtype != self.vae.dtype: + if torch.backends.mps.is_available(): + # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272 + self.vae = self.vae.to(latents.dtype) + + # unscale/denormalize the latents + # denormalize with the mean and std if available and not None + has_latents_mean = hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None + has_latents_std = hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None + if has_latents_mean and has_latents_std: + latents_mean = ( + torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1).to(latents.device, latents.dtype) + ) + latents_std = ( + torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1).to(latents.device, latents.dtype) + ) + latents = latents * latents_std / self.vae.config.scaling_factor + latents_mean + else: + latents = latents / self.vae.config.scaling_factor + + image = self.vae.decode(latents, return_dict=False)[0] + + # cast back to fp16 if needed + if needs_upcasting: + self.vae.to(dtype=torch.float16) + else: + return StableDiffusionXLPipelineOutput(images=latents) + + # apply watermark if available + if self.watermark is not None: + image = self.watermark.apply_watermark(image) + + image = self.image_processor.postprocess(image, output_type=output_type) + + if padding_mask_crop is not None: + image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image] + + # Offload all models + self.maybe_free_model_hooks() + + if not return_dict: + return (image,) + + return StableDiffusionXLPipelineOutput(images=image) From 324db212843c796b79e2cd77aa5b59baeac78316 Mon Sep 17 00:00:00 2001 From: Teriks Date: Mon, 21 Apr 2025 17:48:04 -0500 Subject: [PATCH 2/3] make style --- .../pipeline_controlnet_xl_kolors.py | 58 ++++++------------- .../pipeline_controlnet_xl_kolors_img2img.py | 47 ++++++++------- .../pipeline_controlnet_xl_kolors_inpaint.py | 52 ++++++++--------- .../community/pipeline_kolors_inpainting.py | 25 ++++---- .../loaders/lora_conversion_utils.py | 4 +- .../pipelines/pipeline_flax_utils.py | 2 +- .../pipelines/pipeline_loading_utils.py | 4 +- .../test_stable_diffusion_diffedit.py | 2 +- 8 files changed, 90 insertions(+), 104 deletions(-) diff --git a/examples/community/pipeline_controlnet_xl_kolors.py b/examples/community/pipeline_controlnet_xl_kolors.py index 04b33260f97c..f02ceec1d0e1 100644 --- a/examples/community/pipeline_controlnet_xl_kolors.py +++ b/examples/community/pipeline_controlnet_xl_kolors.py @@ -31,27 +31,31 @@ FromSingleFileMixin, IPAdapterMixin, StableDiffusionXLLoraLoaderMixin, - TextualInversionLoaderMixin + TextualInversionLoaderMixin, +) +from diffusers.models import ( + AutoencoderKL, + ControlNetModel, + ImageProjection, + MultiControlNetModel, + UNet2DConditionModel, ) -from diffusers.models import AutoencoderKL, ImageProjection, UNet2DConditionModel from diffusers.models.attention_processor import ( AttnProcessor2_0, XFormersAttnProcessor, ) +from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer +from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin +from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput from diffusers.schedulers import KarrasDiffusionSchedulers from diffusers.utils import ( - is_invisible_watermark_available, - replace_example_docstring, deprecate, + is_invisible_watermark_available, logging, + replace_example_docstring, ) from diffusers.utils.torch_utils import is_compiled_module, randn_tensor -from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin -from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput -from diffusers.models import ControlNetModel, MultiControlNetModel - -from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer if is_invisible_watermark_available(): from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker @@ -68,33 +72,33 @@ >>> import numpy as np >>> import cv2 >>> from PIL import Image - + >>> prompt = "aerial view, a futuristic research complex in a bright foggy jungle, hard lighting" >>> negative_prompt = "low quality, bad quality, sketches" - + >>> # download an image >>> image = load_image( ... "https://hf.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/hf-logo.png" ... ) - + >>> # initialize the models and pipeline >>> controlnet_conditioning_scale = 0.5 # recommended for good generalization >>> controlnet = ControlNetModel.from_pretrained( ... "Kwai-Kolors/Kolors-ControlNet-Canny", torch_dtype=torch.float16 ... ) - + >>> pipe = KolorsControlNetPipeline.from_pretrained( ... "Kwai-Kolors/Kolors-diffusers", controlnet=controlnet, torch_dtype=torch.float16 ... ) >>> pipe.enable_model_cpu_offload() - + >>> # get canny image >>> image = np.array(image) >>> image = cv2.Canny(image, 100, 200) >>> image = image[:, :, None] >>> image = np.concatenate([image, image, image], axis=2) >>> canny_image = Image.fromarray(image) - + >>> # generate image >>> image = pipe( ... prompt, controlnet_conditioning_scale=controlnet_conditioning_scale, image=canny_image @@ -388,30 +392,6 @@ def encode_prompt( return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds - def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None): - dtype = next(self.image_encoder.parameters()).dtype - - if not isinstance(image, torch.Tensor): - image = self.feature_extractor(image, return_tensors="pt").pixel_values - - image = image.to(device=device, dtype=dtype) - if output_hidden_states: - image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2] - image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) - uncond_image_enc_hidden_states = self.image_encoder( - torch.zeros_like(image), output_hidden_states=True - ).hidden_states[-2] - uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave( - num_images_per_prompt, dim=0 - ) - return image_enc_hidden_states, uncond_image_enc_hidden_states - else: - image_embeds = self.image_encoder(image).image_embeds - image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) - uncond_image_embeds = torch.zeros_like(image_embeds) - - return image_embeds, uncond_image_embeds - def prepare_ip_adapter_image_embeds( self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance ): diff --git a/examples/community/pipeline_controlnet_xl_kolors_img2img.py b/examples/community/pipeline_controlnet_xl_kolors_img2img.py index 29be6d08a428..b9b8abb0d16b 100644 --- a/examples/community/pipeline_controlnet_xl_kolors_img2img.py +++ b/examples/community/pipeline_controlnet_xl_kolors_img2img.py @@ -16,41 +16,46 @@ import inspect from typing import Any, Callable, Dict, List, Optional, Tuple, Union -import PIL.Image import numpy as np +import PIL.Image import torch import torch.nn.functional as F +from transformers import ( + CLIPImageProcessor, + CLIPVisionModelWithProjection, +) + from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback from diffusers.image_processor import PipelineImageInput, VaeImageProcessor from diffusers.loaders import ( FromSingleFileMixin, IPAdapterMixin, StableDiffusionXLLoraLoaderMixin, - TextualInversionLoaderMixin + TextualInversionLoaderMixin, +) +from diffusers.models import ( + AutoencoderKL, + ControlNetModel, + ImageProjection, + MultiControlNetModel, + UNet2DConditionModel, ) -from diffusers.models import AutoencoderKL, ImageProjection, UNet2DConditionModel -from diffusers.models import ControlNetModel, MultiControlNetModel from diffusers.models.attention_processor import ( AttnProcessor2_0, XFormersAttnProcessor, ) - +from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput from diffusers.schedulers import KarrasDiffusionSchedulers from diffusers.utils import ( - is_invisible_watermark_available, - replace_example_docstring, deprecate, + is_invisible_watermark_available, logging, + replace_example_docstring, ) from diffusers.utils.torch_utils import is_compiled_module, randn_tensor -from transformers import ( - CLIPImageProcessor, - CLIPVisionModelWithProjection, -) -from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer if is_invisible_watermark_available(): from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker @@ -64,11 +69,11 @@ >>> import torch >>> import numpy as np >>> from PIL import Image - + >>> from transformers import DPTImageProcessor, DPTForDepthEstimation >>> from diffusers import ControlNetModel, KolorsControlNetImg2ImgPipeline >>> from diffusers.utils import load_image - + >>> depth_estimator = DPTForDepthEstimation.from_pretrained("Intel/dpt-hybrid-midas").to("cuda") >>> feature_extractor = DPTImageProcessor.from_pretrained("Intel/dpt-hybrid-midas") >>> controlnet = ControlNetModel.from_pretrained( @@ -85,14 +90,14 @@ ... torch_dtype=torch.float16, ... ) >>> pipe.enable_model_cpu_offload() - - + + >>> def get_depth_map(image): ... image = feature_extractor(images=image, return_tensors="pt").pixel_values.to("cuda") - ... + ... ... with torch.no_grad(), torch.autocast("cuda"): ... depth_map = depth_estimator(image).predicted_depth - ... + ... ... depth_map = torch.nn.functional.interpolate( ... depth_map.unsqueeze(1), ... size=(1024, 1024), @@ -106,8 +111,8 @@ ... image = image.permute(0, 2, 3, 1).cpu().numpy()[0] ... image = Image.fromarray((image * 255.0).clip(0, 255).astype(np.uint8)) ... return image - - + + >>> prompt = "A robot, 4k photo" >>> image = load_image( ... "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main" @@ -115,7 +120,7 @@ ... ).resize((1024, 1024)) >>> controlnet_conditioning_scale = 0.5 # recommended for good generalization >>> depth_image = get_depth_map(image) - + >>> images = pipe( ... prompt, ... image=image, diff --git a/examples/community/pipeline_controlnet_xl_kolors_inpaint.py b/examples/community/pipeline_controlnet_xl_kolors_inpaint.py index e587b0f3dcd2..b673bf5cbb5e 100644 --- a/examples/community/pipeline_controlnet_xl_kolors_inpaint.py +++ b/examples/community/pipeline_controlnet_xl_kolors_inpaint.py @@ -15,41 +15,41 @@ import inspect from typing import Any, Callable, Dict, List, Optional, Tuple, Union -import PIL.Image import numpy as np +import PIL.Image import torch import torch.nn.functional as F +from transformers import ( + CLIPImageProcessor, + CLIPVisionModelWithProjection, +) + from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback from diffusers.image_processor import PipelineImageInput, VaeImageProcessor from diffusers.loaders import ( FromSingleFileMixin, IPAdapterMixin, StableDiffusionXLLoraLoaderMixin, - TextualInversionLoaderMixin + TextualInversionLoaderMixin, +) +from diffusers.models import ( + AutoencoderKL, + ControlNetModel, + ImageProjection, + MultiControlNetModel, + UNet2DConditionModel, ) -from diffusers.models import AutoencoderKL, ImageProjection, UNet2DConditionModel -from diffusers.models import ControlNetModel, MultiControlNetModel from diffusers.models.attention_processor import ( AttnProcessor2_0, XFormersAttnProcessor, ) - +from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput from diffusers.schedulers import KarrasDiffusionSchedulers -from diffusers.utils import ( - is_invisible_watermark_available, - replace_example_docstring, - deprecate, - logging -) +from diffusers.utils import deprecate, is_invisible_watermark_available, logging, replace_example_docstring from diffusers.utils.torch_utils import is_compiled_module, randn_tensor -from transformers import ( - CLIPImageProcessor, - CLIPVisionModelWithProjection, -) -from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer if is_invisible_watermark_available(): from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker @@ -66,20 +66,20 @@ >>> import numpy as np >>> import torch >>> import cv2 - + >>> init_image = load_image( ... "https://huggingface.co/datasets/diffusers/test-arrays/resolve/main/stable_diffusion_inpaint/boy.png" ... ) >>> init_image = init_image.resize((1024, 1024)) - + >>> generator = torch.Generator(device="cpu").manual_seed(1) - + >>> mask_image = load_image( ... "https://huggingface.co/datasets/diffusers/test-arrays/resolve/main/stable_diffusion_inpaint/boy_mask.png" ... ) >>> mask_image = mask_image.resize((1024, 1024)) - - + + >>> def make_canny_condition(image): ... image = np.array(image) ... image = cv2.Canny(image, 100, 200) @@ -87,19 +87,19 @@ ... image = np.concatenate([image, image, image], axis=2) ... image = Image.fromarray(image) ... return image - - + + >>> control_image = make_canny_condition(init_image) - + >>> controlnet = ControlNetModel.from_pretrained( ... "Kwai-Kolors/Kolors-ControlNet-Canny", torch_dtype=torch.float16 ... ) >>> pipe = KolorsControlNetInpaintPipeline.from_pretrained( ... "Kwai-Kolors/Kolors-diffusers", controlnet=controlnet, torch_dtype=torch.float16 ... ) - + >>> pipe.enable_model_cpu_offload() - + # generate image >>> image = pipe( ... "a handsome man with ray-ban sunglasses", diff --git a/examples/community/pipeline_kolors_inpainting.py b/examples/community/pipeline_kolors_inpainting.py index aba8535da25d..d6b83c0a1825 100644 --- a/examples/community/pipeline_kolors_inpainting.py +++ b/examples/community/pipeline_kolors_inpainting.py @@ -15,9 +15,14 @@ import inspect from typing import Any, Callable, Dict, List, Optional, Tuple, Union -import PIL.Image import numpy as np +import PIL.Image import torch +from transformers import ( + CLIPImageProcessor, + CLIPVisionModelWithProjection, +) + from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback from diffusers.image_processor import PipelineImageInput, VaeImageProcessor from diffusers.loaders import ( @@ -33,23 +38,19 @@ LoRAXFormersAttnProcessor, XFormersAttnProcessor, ) +from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput from diffusers.schedulers import KarrasDiffusionSchedulers from diffusers.utils import ( + deprecate, is_invisible_watermark_available, - replace_example_docstring, is_torch_xla_available, - deprecate, logging, + replace_example_docstring, ) from diffusers.utils.torch_utils import randn_tensor -from transformers import ( - CLIPImageProcessor, - CLIPVisionModelWithProjection, -) -from diffusers.pipelines.kolors import ChatGLMModel, ChatGLMTokenizer if is_invisible_watermark_available(): from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker @@ -70,20 +71,20 @@ >>> import torch >>> from diffusers import KolorsInpaintPipeline >>> from diffusers.utils import load_image - + >>> pipe = KolorsInpaintPipeline.from_pretrained( ... "Kwai-Kolors/Kolors-diffusers", ... torch_dtype=torch.float16, ... variant="fp16" ... ) >>> pipe.to("cuda") - + >>> img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png" >>> mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png" - + >>> init_image = load_image(img_url).convert("RGB") >>> mask_image = load_image(mask_url).convert("RGB") - + >>> prompt = "A majestic tiger sitting on a bench" >>> image = pipe( ... prompt=prompt, image=init_image, mask_image=mask_image, num_inference_steps=50, strength=0.80 diff --git a/src/diffusers/loaders/lora_conversion_utils.py b/src/diffusers/loaders/lora_conversion_utils.py index 7fec3299eeac..d0c9611735ce 100644 --- a/src/diffusers/loaders/lora_conversion_utils.py +++ b/src/diffusers/loaders/lora_conversion_utils.py @@ -433,7 +433,7 @@ def _convert_to_ai_toolkit_cat(sds_sd, ait_sd, sds_key, ait_keys, dims=None): ait_up_keys = [k + ".lora_B.weight" for k in ait_keys] if not is_sparse: # down_weight is copied to each split - ait_sd.update({k: down_weight for k in ait_down_keys}) + ait_sd.update(dict.fromkeys(ait_down_keys, down_weight)) # up_weight is split to each split ait_sd.update({k: v for k, v in zip(ait_up_keys, torch.split(up_weight, dims, dim=0))}) # noqa: C416 @@ -923,7 +923,7 @@ def handle_qkv(sds_sd, ait_sd, sds_key, ait_keys, dims=None): ait_up_keys = [k + ".lora_B.weight" for k in ait_keys] # down_weight is copied to each split - ait_sd.update({k: down_weight for k in ait_down_keys}) + ait_sd.update(dict.fromkeys(ait_down_keys, down_weight)) # up_weight is split to each split ait_sd.update({k: v for k, v in zip(ait_up_keys, torch.split(up_weight, dims, dim=0))}) # noqa: C416 diff --git a/src/diffusers/pipelines/pipeline_flax_utils.py b/src/diffusers/pipelines/pipeline_flax_utils.py index ec2f82bcf742..54ab7d19e3fb 100644 --- a/src/diffusers/pipelines/pipeline_flax_utils.py +++ b/src/diffusers/pipelines/pipeline_flax_utils.py @@ -469,7 +469,7 @@ def load_module(name, value): class_obj = import_flax_or_no_model(pipeline_module, class_name) importable_classes = ALL_IMPORTABLE_CLASSES - class_candidates = {c: class_obj for c in importable_classes.keys()} + class_candidates = dict.fromkeys(importable_classes.keys(), class_obj) else: # else we just import it from the library. library = importlib.import_module(library_name) diff --git a/src/diffusers/pipelines/pipeline_loading_utils.py b/src/diffusers/pipelines/pipeline_loading_utils.py index 89a403df8d65..9788d758e9bc 100644 --- a/src/diffusers/pipelines/pipeline_loading_utils.py +++ b/src/diffusers/pipelines/pipeline_loading_utils.py @@ -341,13 +341,13 @@ def get_class_obj_and_candidates( pipeline_module = getattr(pipelines, library_name) class_obj = getattr(pipeline_module, class_name) - class_candidates = {c: class_obj for c in importable_classes.keys()} + class_candidates = dict.fromkeys(importable_classes.keys(), class_obj) elif os.path.isfile(os.path.join(component_folder, library_name + ".py")): # load custom component class_obj = get_class_from_dynamic_module( component_folder, module_file=library_name + ".py", class_name=class_name ) - class_candidates = {c: class_obj for c in importable_classes.keys()} + class_candidates = dict.fromkeys(importable_classes.keys(), class_obj) else: # else we just import it from the library. library = importlib.import_module(library_name) diff --git a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_diffedit.py b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_diffedit.py index 34ea56664a95..c190a789b15b 100644 --- a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_diffedit.py +++ b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_diffedit.py @@ -205,7 +205,7 @@ def test_save_load_optional_components(self): # set all optional components to None and update pipeline config accordingly for optional_component in pipe._optional_components: setattr(pipe, optional_component, None) - pipe.register_modules(**{optional_component: None for optional_component in pipe._optional_components}) + pipe.register_modules(**dict.fromkeys(pipe._optional_components)) inputs = self.get_dummy_inputs(torch_device) output = pipe(**inputs)[0] From 671458f55a16a207080b487fac0f24f7d34ac840 Mon Sep 17 00:00:00 2001 From: Teriks Date: Wed, 23 Apr 2025 00:57:20 -0500 Subject: [PATCH 3/3] Kolors additional community pipelines fixes & SDXL pipeline feature parity Example string doc fixes, make sure variant=fp16 Fix device mismatch for encoder_hidden_states in ControlNetModel patch when sequential offload is enabled. Fix _get_add_time_ids implementations and add proper callback_on_step_end implementations In KolorsControlNetImg2ImgPipeline & KolorsControlNetInpaintPipeline Properly implement __call__ arguments: negative_original_size negative_crops_coords_top_left negative_target_size aesthetic_score negative_aesthetic_score In KolorsControlNetPipeline Properly implement __call__ arguments: negative_original_size negative_crops_coords_top_left negative_target_size This covers all typical SDXL conditioning arguments KolorsControlNetPipeline.__call__ argument "control_image" rename to -> "image" in order to match StableDiffusionXLControlNetPipeline --- .../pipeline_controlnet_xl_kolors.py | 335 ++++++++------ .../pipeline_controlnet_xl_kolors_img2img.py | 391 +++++++++++------ .../pipeline_controlnet_xl_kolors_inpaint.py | 412 ++++++++++-------- .../community/pipeline_kolors_inpainting.py | 289 ++++++------ 4 files changed, 835 insertions(+), 592 deletions(-) diff --git a/examples/community/pipeline_controlnet_xl_kolors.py b/examples/community/pipeline_controlnet_xl_kolors.py index f02ceec1d0e1..b805c9a04a07 100644 --- a/examples/community/pipeline_controlnet_xl_kolors.py +++ b/examples/community/pipeline_controlnet_xl_kolors.py @@ -84,11 +84,17 @@ >>> # initialize the models and pipeline >>> controlnet_conditioning_scale = 0.5 # recommended for good generalization >>> controlnet = ControlNetModel.from_pretrained( - ... "Kwai-Kolors/Kolors-ControlNet-Canny", torch_dtype=torch.float16 + ... "Kwai-Kolors/Kolors-ControlNet-Canny", + ... use_safetensors=True, + ... torch_dtype=torch.float16 ... ) >>> pipe = KolorsControlNetPipeline.from_pretrained( - ... "Kwai-Kolors/Kolors-diffusers", controlnet=controlnet, torch_dtype=torch.float16 + ... "Kwai-Kolors/Kolors-diffusers", + ... controlnet=controlnet, + ... variant="fp16", + ... use_safetensors=True, + ... torch_dtype=torch.float16 ... ) >>> pipe.enable_model_cpu_offload() @@ -109,7 +115,7 @@ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents def retrieve_latents( - encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" + encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" ): if hasattr(encoder_output, "latent_dist") and sample_mode == "sample": return encoder_output.latent_dist.sample(generator) @@ -180,22 +186,23 @@ class KolorsControlNetPipeline( "add_text_embeds", "add_time_ids", "negative_pooled_prompt_embeds", - "add_neg_time_ids", + "negative_add_time_ids", + "image", ] def __init__( - self, - vae: AutoencoderKL, - text_encoder: ChatGLMModel, - tokenizer: ChatGLMTokenizer, - unet: UNet2DConditionModel, - controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel], - scheduler: KarrasDiffusionSchedulers, - requires_aesthetics_score: bool = False, - force_zeros_for_empty_prompt: bool = True, - feature_extractor: CLIPImageProcessor = None, - image_encoder: CLIPVisionModelWithProjection = None, - add_watermarker: Optional[bool] = None, + self, + vae: AutoencoderKL, + text_encoder: ChatGLMModel, + tokenizer: ChatGLMTokenizer, + unet: UNet2DConditionModel, + controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel], + scheduler: KarrasDiffusionSchedulers, + requires_aesthetics_score: bool = False, + force_zeros_for_empty_prompt: bool = True, + feature_extractor: CLIPImageProcessor = None, + image_encoder: CLIPVisionModelWithProjection = None, + add_watermarker: Optional[bool] = None, ): super().__init__() @@ -227,17 +234,17 @@ def __init__( self.register_to_config(requires_aesthetics_score=requires_aesthetics_score) def encode_prompt( - self, - prompt, - device: Optional[torch.device] = None, - num_images_per_prompt: int = 1, - do_classifier_free_guidance: bool = True, - negative_prompt=None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - lora_scale: Optional[float] = None, + self, + prompt, + device: Optional[torch.device] = None, + num_images_per_prompt: int = 1, + do_classifier_free_guidance: bool = True, + negative_prompt=None, + prompt_embeds: Optional[torch.FloatTensor] = None, + negative_prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + lora_scale: Optional[float] = None, ): r""" Encodes the prompt into text encoder hidden states. @@ -305,10 +312,11 @@ def encode_prompt( return_tensors="pt", ).to(self._execution_device) output = text_encoder( - input_ids=text_inputs['input_ids'], - attention_mask=text_inputs['attention_mask'], - position_ids=text_inputs['position_ids'], - output_hidden_states=True) + input_ids=text_inputs["input_ids"], + attention_mask=text_inputs["attention_mask"], + position_ids=text_inputs["position_ids"], + output_hidden_states=True, + ) prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] bs_embed, seq_len, _ = prompt_embeds.shape @@ -359,10 +367,11 @@ def encode_prompt( return_tensors="pt", ).to(self._execution_device) output = text_encoder( - input_ids=uncond_input['input_ids'], - attention_mask=uncond_input['attention_mask'], - position_ids=uncond_input['position_ids'], - output_hidden_states=True) + input_ids=uncond_input["input_ids"], + attention_mask=uncond_input["attention_mask"], + position_ids=uncond_input["position_ids"], + output_hidden_states=True, + ) negative_prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() negative_pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] @@ -393,7 +402,7 @@ def encode_prompt( return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds def prepare_ip_adapter_image_embeds( - self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance + self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance ): image_embeds = [] if do_classifier_free_guidance: @@ -409,7 +418,7 @@ def prepare_ip_adapter_image_embeds( ) for single_ip_adapter_image, image_proj_layer in zip( - ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers + ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers ): output_hidden_state = not isinstance(image_proj_layer, ImageProjection) single_image_embeds, single_negative_image_embeds = self.encode_image( @@ -468,22 +477,22 @@ def prepare_extra_step_kwargs(self, generator, eta): return extra_step_kwargs def check_inputs( - self, - prompt, - image, - num_inference_steps, - callback_steps, - negative_prompt=None, - prompt_embeds=None, - negative_prompt_embeds=None, - pooled_prompt_embeds=None, - negative_pooled_prompt_embeds=None, - ip_adapter_image=None, - ip_adapter_image_embeds=None, - controlnet_conditioning_scale=1.0, - control_guidance_start=0.0, - control_guidance_end=1.0, - callback_on_step_end_tensor_inputs=None, + self, + prompt, + image, + num_inference_steps, + callback_steps, + negative_prompt=None, + prompt_embeds=None, + negative_prompt_embeds=None, + pooled_prompt_embeds=None, + negative_pooled_prompt_embeds=None, + ip_adapter_image=None, + ip_adapter_image_embeds=None, + controlnet_conditioning_scale=1.0, + control_guidance_start=0.0, + control_guidance_end=1.0, + callback_on_step_end_tensor_inputs=None, ): if num_inference_steps is None: raise ValueError("`num_inference_steps` cannot be None.") @@ -500,7 +509,7 @@ def check_inputs( ) if callback_on_step_end_tensor_inputs is not None and not all( - k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs + k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs ): raise ValueError( f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" @@ -557,22 +566,22 @@ def check_inputs( # Check `controlnet_conditioning_scale` if ( - isinstance(self.controlnet, ControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, ControlNetModel) + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) ): if not isinstance(controlnet_conditioning_scale, float): raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.") elif ( - isinstance(self.controlnet, MultiControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) ): if isinstance(controlnet_conditioning_scale, list): if any(isinstance(i, list) for i in controlnet_conditioning_scale): raise ValueError("A single batch of multiple conditionings are supported at the moment.") elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len( - self.controlnet.nets + self.controlnet.nets ): raise ValueError( "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have" @@ -633,12 +642,12 @@ def check_image(self, image, prompt, prompt_embeds): image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray) if ( - not image_is_pil - and not image_is_tensor - and not image_is_np - and not image_is_pil_list - and not image_is_tensor_list - and not image_is_np_list + not image_is_pil + and not image_is_tensor + and not image_is_np + and not image_is_pil_list + and not image_is_tensor_list + and not image_is_np_list ): raise TypeError( f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}" @@ -663,16 +672,16 @@ def check_image(self, image, prompt, prompt_embeds): # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.prepare_image def prepare_control_image( - self, - image, - width, - height, - batch_size, - num_images_per_prompt, - device, - dtype, - do_classifier_free_guidance=False, - guess_mode=False, + self, + image, + width, + height, + batch_size, + num_images_per_prompt, + device, + dtype, + do_classifier_free_guidance=False, + guess_mode=False, ): image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) image_batch_size = image.shape[0] @@ -711,8 +720,9 @@ def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype return latents # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents - def prepare_latents_t2i(self, batch_size, num_channels_latents, height, width, dtype, device, generator, - latents=None): + def prepare_latents_t2i( + self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None + ): shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) if isinstance(generator, list) and len(generator) != batch_size: raise ValueError( @@ -729,11 +739,14 @@ def prepare_latents_t2i(self, batch_size, num_channels_latents, height, width, d latents = latents * self.scheduler.init_noise_sigma return latents - def _get_add_time_ids(self, original_size, crops_coords_top_left, target_size, dtype): + # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline._get_add_time_ids + def _get_add_time_ids( + self, original_size, crops_coords_top_left, target_size, dtype, text_encoder_projection_dim=None + ): add_time_ids = list(original_size + crops_coords_top_left + target_size) passed_add_embed_dim = ( - self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 + self.unet.config.addition_time_embed_dim * len(add_time_ids) + text_encoder_projection_dim ) expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features @@ -785,39 +798,42 @@ def num_timesteps(self): @torch.no_grad() @replace_example_docstring(EXAMPLE_DOC_STRING) def __call__( - self, - prompt: Union[str, List[str]] = None, - control_image: PipelineImageInput = None, - height: Optional[int] = None, - width: Optional[int] = None, - num_inference_steps: int = 50, - guidance_scale: float = 5.0, - negative_prompt: Optional[Union[str, List[str]]] = None, - num_images_per_prompt: Optional[int] = 1, - eta: float = 0.0, - guess_mode: bool = False, - generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, - latents: Optional[torch.Tensor] = None, - prompt_embeds: Optional[torch.Tensor] = None, - negative_prompt_embeds: Optional[torch.Tensor] = None, - pooled_prompt_embeds: Optional[torch.Tensor] = None, - negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, - ip_adapter_image: Optional[PipelineImageInput] = None, - ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, - output_type: Optional[str] = "pil", - return_dict: bool = True, - cross_attention_kwargs: Optional[Dict[str, Any]] = None, - controlnet_conditioning_scale: Union[float, List[float]] = 0.8, - control_guidance_start: Union[float, List[float]] = 0.0, - control_guidance_end: Union[float, List[float]] = 1.0, - original_size: Tuple[int, int] = None, - crops_coords_top_left: Tuple[int, int] = (0, 0), - target_size: Tuple[int, int] = None, - callback_on_step_end: Optional[ - Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] - ] = None, - callback_on_step_end_tensor_inputs: List[str] = ["latents"], - **kwargs, + self, + prompt: Union[str, List[str]] = None, + image: PipelineImageInput = None, + height: Optional[int] = None, + width: Optional[int] = None, + num_inference_steps: int = 50, + guidance_scale: float = 5.0, + negative_prompt: Optional[Union[str, List[str]]] = None, + num_images_per_prompt: Optional[int] = 1, + eta: float = 0.0, + guess_mode: bool = False, + generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, + latents: Optional[torch.Tensor] = None, + prompt_embeds: Optional[torch.Tensor] = None, + negative_prompt_embeds: Optional[torch.Tensor] = None, + pooled_prompt_embeds: Optional[torch.Tensor] = None, + negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, + ip_adapter_image: Optional[PipelineImageInput] = None, + ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, + output_type: Optional[str] = "pil", + return_dict: bool = True, + cross_attention_kwargs: Optional[Dict[str, Any]] = None, + controlnet_conditioning_scale: Union[float, List[float]] = 0.8, + control_guidance_start: Union[float, List[float]] = 0.0, + control_guidance_end: Union[float, List[float]] = 1.0, + original_size: Tuple[int, int] = None, + crops_coords_top_left: Tuple[int, int] = (0, 0), + target_size: Tuple[int, int] = None, + negative_original_size: Optional[Tuple[int, int]] = None, + negative_crops_coords_top_left: Tuple[int, int] = (0, 0), + negative_target_size: Optional[Tuple[int, int]] = None, + callback_on_step_end: Optional[ + Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] + ] = None, + callback_on_step_end_tensor_inputs: List[str] = ["latents"], + **kwargs, ): r""" Function invoked when calling the pipeline for generation. @@ -826,7 +842,7 @@ def __call__( prompt (`str` or `List[str]`, *optional*): The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`. instead. - control_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: + image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): The ControlNet input condition. ControlNet uses this input condition to generate guidance to Unet. If the type is specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also @@ -834,11 +850,11 @@ def __call__( and/or width are passed, `image` is resized according to them. If multiple ControlNets are specified in init, images must be passed as a list such that each element of the list can be correctly batched for input to a single controlnet. - height (`int`, *optional*, defaults to the size of control_image): + height (`int`, *optional*, defaults to the size of image): The height in pixels of the generated image. Anything below 512 pixels won't work well for [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) and checkpoints that are not specifically fine-tuned on low resolutions. - width (`int`, *optional*, defaults to the size of control_image): + width (`int`, *optional*, defaults to the size of image): The width in pixels of the generated image. Anything below 512 pixels won't work well for [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) and checkpoints that are not specifically fine-tuned on low resolutions. @@ -925,6 +941,21 @@ def __call__( For most cases, `target_size` should be set to the desired height and width of the generated image. If not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + To negatively condition the generation process based on a specific image resolution. Part of SDXL's + micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): + To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's + micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + To negatively condition the generation process based on a target image resolution. It should be as same + as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*): A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of each denoising step during the inference. with the following arguments: `callback_on_step_end(self: @@ -980,7 +1011,7 @@ def __call__( # 1. Check inputs. Raise error if not correct self.check_inputs( prompt, - control_image, + image, num_inference_steps, callback_steps, negative_prompt, @@ -1045,8 +1076,8 @@ def __call__( ) if isinstance(controlnet, ControlNetModel): - control_image = self.prepare_control_image( - image=control_image, + image = self.prepare_control_image( + image=image, width=width, height=height, batch_size=batch_size * num_images_per_prompt, @@ -1056,11 +1087,11 @@ def __call__( do_classifier_free_guidance=self.do_classifier_free_guidance, guess_mode=guess_mode, ) - height, width = control_image.shape[-2:] + height, width = image.shape[-2:] elif isinstance(controlnet, MultiControlNetModel): control_images = [] - for control_image_ in control_image: + for control_image_ in image: control_image_ = self.prepare_control_image( image=control_image_, width=width, @@ -1075,8 +1106,8 @@ def __call__( control_images.append(control_image_) - control_image = control_images - height, width = control_image[0].shape[-2:] + image = control_images + height, width = image[0].shape[-2:] else: assert False @@ -1119,18 +1150,35 @@ def __call__( controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps) # 7.2 Prepare added time ids & embeddings - if isinstance(control_image, list): - original_size = original_size or control_image[0].shape[-2:] + if isinstance(image, list): + original_size = original_size or image[0].shape[-2:] else: - original_size = original_size or control_image.shape[-2:] + original_size = original_size or image.shape[-2:] target_size = target_size or (height, width) # 7. Prepare added time ids & embeddings + text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1]) + add_text_embeds = pooled_prompt_embeds add_time_ids = self._get_add_time_ids( - original_size, crops_coords_top_left, target_size, dtype=prompt_embeds.dtype + original_size, + crops_coords_top_left, + target_size, + dtype=prompt_embeds.dtype, + text_encoder_projection_dim=text_encoder_projection_dim, ) + if negative_original_size is not None and negative_target_size is not None: + negative_add_time_ids = self._get_add_time_ids( + negative_original_size, + negative_crops_coords_top_left, + negative_target_size, + dtype=prompt_embeds.dtype, + text_encoder_projection_dim=text_encoder_projection_dim, + ) + else: + negative_add_time_ids = add_time_ids + if self.do_classifier_free_guidance: prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0) @@ -1153,10 +1201,12 @@ def __call__( cn_og_forward = cn_model.forward def _cn_patch_forward(*args, **kwargs): - encoder_hidden_states = kwargs['encoder_hidden_states'] + encoder_hidden_states = kwargs["encoder_hidden_states"] if cn_model.encoder_hid_proj is not None and cn_model.config.encoder_hid_dim_type == "text_proj": - encoder_hidden_states = cn_model.encoder_hid_proj(kwargs['encoder_hidden_states']) - kwargs.pop('encoder_hidden_states') + # Ensure encoder_hidden_states is on the same device as the projection layer + encoder_hidden_states = encoder_hidden_states.to(cn_model.encoder_hid_proj.weight.device) + encoder_hidden_states = cn_model.encoder_hid_proj(encoder_hidden_states) + kwargs.pop("encoder_hidden_states") return cn_og_forward(*args, encoder_hidden_states=encoder_hidden_states, **kwargs) cn_model.forward = _cn_patch_forward @@ -1200,7 +1250,7 @@ def _cn_patch_forward(*args, **kwargs): control_model_input, t, encoder_hidden_states=controlnet_prompt_embeds, - controlnet_cond=control_image, + controlnet_cond=image, conditioning_scale=cond_scale, guess_mode=guess_mode, added_cond_kwargs=controlnet_added_cond_kwargs, @@ -1212,7 +1262,9 @@ def _cn_patch_forward(*args, **kwargs): # To apply the output of ControlNet to both the unconditional and conditional batches, # add 0 to the unconditional batch to keep it unchanged. down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples] - mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample]) + mid_block_res_sample = torch.cat( + [torch.zeros_like(mid_block_res_sample), mid_block_res_sample] + ) if ip_adapter_image is not None or ip_adapter_image_embeds is not None: added_cond_kwargs["image_embeds"] = image_embeds @@ -1238,6 +1290,23 @@ def _cn_patch_forward(*args, **kwargs): # compute the previous noisy sample x_t -> x_t-1 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] + if callback_on_step_end is not None: + callback_kwargs = {} + for k in callback_on_step_end_tensor_inputs: + callback_kwargs[k] = locals()[k] + callback_outputs = callback_on_step_end(self, i, t, callback_kwargs) + + latents = callback_outputs.pop("latents", latents) + prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) + negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds) + add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds) + negative_pooled_prompt_embeds = callback_outputs.pop( + "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds + ) + add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids) + negative_add_time_ids = callback_outputs.pop("negative_add_time_ids", negative_add_time_ids) + image = callback_outputs.pop("image", image) + # call the callback, if provided if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): progress_bar.update() diff --git a/examples/community/pipeline_controlnet_xl_kolors_img2img.py b/examples/community/pipeline_controlnet_xl_kolors_img2img.py index b9b8abb0d16b..5cfb98d9694f 100644 --- a/examples/community/pipeline_controlnet_xl_kolors_img2img.py +++ b/examples/community/pipeline_controlnet_xl_kolors_img2img.py @@ -78,16 +78,15 @@ >>> feature_extractor = DPTImageProcessor.from_pretrained("Intel/dpt-hybrid-midas") >>> controlnet = ControlNetModel.from_pretrained( ... "Kwai-Kolors/Kolors-ControlNet-Depth", - ... variant="fp16", ... use_safetensors=True, - ... torch_dtype=torch.float16, + ... torch_dtype=torch.float16 ... ) >>> pipe = KolorsControlNetImg2ImgPipeline.from_pretrained( ... "Kwai-Kolors/Kolors-diffusers", ... controlnet=controlnet, ... variant="fp16", ... use_safetensors=True, - ... torch_dtype=torch.float16, + ... torch_dtype=torch.float16 ... ) >>> pipe.enable_model_cpu_offload() @@ -125,7 +124,7 @@ ... prompt, ... image=image, ... control_image=depth_image, - ... strength=0.99, + ... strength=0.80, ... num_inference_steps=50, ... controlnet_conditioning_scale=controlnet_conditioning_scale, ... ).images @@ -136,7 +135,7 @@ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents def retrieve_latents( - encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" + encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" ): if hasattr(encoder_output, "latent_dist") and sample_mode == "sample": return encoder_output.latent_dist.sample(generator) @@ -209,21 +208,22 @@ class KolorsControlNetImg2ImgPipeline( "add_time_ids", "negative_pooled_prompt_embeds", "add_neg_time_ids", + "control_image", ] def __init__( - self, - vae: AutoencoderKL, - text_encoder: ChatGLMModel, - tokenizer: ChatGLMTokenizer, - unet: UNet2DConditionModel, - controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel], - scheduler: KarrasDiffusionSchedulers, - requires_aesthetics_score: bool = False, - force_zeros_for_empty_prompt: bool = True, - feature_extractor: CLIPImageProcessor = None, - image_encoder: CLIPVisionModelWithProjection = None, - add_watermarker: Optional[bool] = None, + self, + vae: AutoencoderKL, + text_encoder: ChatGLMModel, + tokenizer: ChatGLMTokenizer, + unet: UNet2DConditionModel, + controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel], + scheduler: KarrasDiffusionSchedulers, + requires_aesthetics_score: bool = False, + force_zeros_for_empty_prompt: bool = True, + feature_extractor: CLIPImageProcessor = None, + image_encoder: CLIPVisionModelWithProjection = None, + add_watermarker: Optional[bool] = None, ): super().__init__() @@ -255,17 +255,17 @@ def __init__( self.register_to_config(requires_aesthetics_score=requires_aesthetics_score) def encode_prompt( - self, - prompt, - device: Optional[torch.device] = None, - num_images_per_prompt: int = 1, - do_classifier_free_guidance: bool = True, - negative_prompt=None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - lora_scale: Optional[float] = None, + self, + prompt, + device: Optional[torch.device] = None, + num_images_per_prompt: int = 1, + do_classifier_free_guidance: bool = True, + negative_prompt=None, + prompt_embeds: Optional[torch.FloatTensor] = None, + negative_prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + lora_scale: Optional[float] = None, ): r""" Encodes the prompt into text encoder hidden states. @@ -334,10 +334,11 @@ def encode_prompt( return_tensors="pt", ).to(self._execution_device) output = text_encoder( - input_ids=text_inputs['input_ids'], - attention_mask=text_inputs['attention_mask'], - position_ids=text_inputs['position_ids'], - output_hidden_states=True) + input_ids=text_inputs["input_ids"], + attention_mask=text_inputs["attention_mask"], + position_ids=text_inputs["position_ids"], + output_hidden_states=True, + ) prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] bs_embed, seq_len, _ = prompt_embeds.shape @@ -390,10 +391,11 @@ def encode_prompt( return_tensors="pt", ).to(self._execution_device) output = text_encoder( - input_ids=uncond_input['input_ids'], - attention_mask=uncond_input['attention_mask'], - position_ids=uncond_input['position_ids'], - output_hidden_states=True) + input_ids=uncond_input["input_ids"], + attention_mask=uncond_input["attention_mask"], + position_ids=uncond_input["position_ids"], + output_hidden_states=True, + ) negative_prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() negative_pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] @@ -455,7 +457,7 @@ def encode_image(self, image, device, num_images_per_prompt, output_hidden_state # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds def prepare_ip_adapter_image_embeds( - self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance + self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance ): image_embeds = [] if do_classifier_free_guidance: @@ -470,7 +472,7 @@ def prepare_ip_adapter_image_embeds( ) for single_ip_adapter_image, image_proj_layer in zip( - ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers + ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers ): output_hidden_state = not isinstance(image_proj_layer, ImageProjection) single_image_embeds, single_negative_image_embeds = self.encode_image( @@ -502,7 +504,7 @@ def prepare_ip_adapter_image_embeds( # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs def prepare_extra_step_kwargs(self, generator, eta): # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature - # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. + # eta (η) is only used with the DDIMScheduler, it will be ignored for others. # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 # and should be between [0, 1] @@ -518,23 +520,23 @@ def prepare_extra_step_kwargs(self, generator, eta): return extra_step_kwargs def check_inputs( - self, - prompt, - image, - strength, - num_inference_steps, - callback_steps, - negative_prompt=None, - prompt_embeds=None, - negative_prompt_embeds=None, - pooled_prompt_embeds=None, - negative_pooled_prompt_embeds=None, - ip_adapter_image=None, - ip_adapter_image_embeds=None, - controlnet_conditioning_scale=1.0, - control_guidance_start=0.0, - control_guidance_end=1.0, - callback_on_step_end_tensor_inputs=None, + self, + prompt, + image, + strength, + num_inference_steps, + callback_steps, + negative_prompt=None, + prompt_embeds=None, + negative_prompt_embeds=None, + pooled_prompt_embeds=None, + negative_pooled_prompt_embeds=None, + ip_adapter_image=None, + ip_adapter_image_embeds=None, + controlnet_conditioning_scale=1.0, + control_guidance_start=0.0, + control_guidance_end=1.0, + callback_on_step_end_tensor_inputs=None, ): if strength < 0 or strength > 1: raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}") @@ -553,7 +555,7 @@ def check_inputs( ) if callback_on_step_end_tensor_inputs is not None and not all( - k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs + k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs ): raise ValueError( f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" @@ -610,15 +612,15 @@ def check_inputs( ) if ( - isinstance(self.controlnet, ControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, ControlNetModel) + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) ): self.check_image(image, prompt, prompt_embeds) elif ( - isinstance(self.controlnet, MultiControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) ): if not isinstance(image, list): raise TypeError("For multiple controlnets: `image` must be type `list`") @@ -639,22 +641,22 @@ def check_inputs( # Check `controlnet_conditioning_scale` if ( - isinstance(self.controlnet, ControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, ControlNetModel) + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) ): if not isinstance(controlnet_conditioning_scale, float): raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.") elif ( - isinstance(self.controlnet, MultiControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) ): if isinstance(controlnet_conditioning_scale, list): if any(isinstance(i, list) for i in controlnet_conditioning_scale): raise ValueError("A single batch of multiple conditionings are supported at the moment.") elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len( - self.controlnet.nets + self.controlnet.nets ): raise ValueError( "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have" @@ -715,12 +717,12 @@ def check_image(self, image, prompt, prompt_embeds): image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray) if ( - not image_is_pil - and not image_is_tensor - and not image_is_np - and not image_is_pil_list - and not image_is_tensor_list - and not image_is_np_list + not image_is_pil + and not image_is_tensor + and not image_is_np + and not image_is_pil_list + and not image_is_tensor_list + and not image_is_np_list ): raise TypeError( f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}" @@ -745,16 +747,16 @@ def check_image(self, image, prompt, prompt_embeds): # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.prepare_image def prepare_control_image( - self, - image, - width, - height, - batch_size, - num_images_per_prompt, - device, - dtype, - do_classifier_free_guidance=False, - guess_mode=False, + self, + image, + width, + height, + batch_size, + num_images_per_prompt, + device, + dtype, + do_classifier_free_guidance=False, + guess_mode=False, ): image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) image_batch_size = image.shape[0] @@ -780,7 +782,7 @@ def get_timesteps(self, num_inference_steps, strength, device): init_timestep = min(int(num_inference_steps * strength), num_inference_steps) t_start = max(num_inference_steps - init_timestep, 0) - timesteps = self.scheduler.timesteps[t_start * self.scheduler.order:] + timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :] if hasattr(self.scheduler, "set_begin_index"): self.scheduler.set_begin_index(t_start * self.scheduler.order) @@ -788,7 +790,7 @@ def get_timesteps(self, num_inference_steps, strength, device): # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.prepare_latents def prepare_latents( - self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True + self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True ): if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)): raise ValueError( @@ -821,7 +823,7 @@ def prepare_latents( elif isinstance(generator, list): init_latents = [ - retrieve_latents(self.vae.encode(image[i: i + 1]), generator=generator[i]) + retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i]) for i in range(batch_size) ] init_latents = torch.cat(init_latents, dim=0) @@ -857,8 +859,9 @@ def prepare_latents( return latents # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents - def prepare_latents_t2i(self, batch_size, num_channels_latents, height, width, dtype, device, generator, - latents=None): + def prepare_latents_t2i( + self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None + ): shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) if isinstance(generator, list) and len(generator) != batch_size: raise ValueError( @@ -875,21 +878,55 @@ def prepare_latents_t2i(self, batch_size, num_channels_latents, height, width, d latents = latents * self.scheduler.init_noise_sigma return latents - def _get_add_time_ids(self, original_size, crops_coords_top_left, target_size, dtype): - add_time_ids = list(original_size + crops_coords_top_left + target_size) + # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline._get_add_time_ids + def _get_add_time_ids( + self, + original_size, + crops_coords_top_left, + target_size, + aesthetic_score, + negative_aesthetic_score, + negative_original_size, + negative_crops_coords_top_left, + negative_target_size, + dtype, + text_encoder_projection_dim=None, + ): + if self.config.requires_aesthetics_score: + add_time_ids = list(original_size + crops_coords_top_left + (aesthetic_score,)) + add_neg_time_ids = list( + negative_original_size + negative_crops_coords_top_left + (negative_aesthetic_score,) + ) + else: + add_time_ids = list(original_size + crops_coords_top_left + target_size) + add_neg_time_ids = list(negative_original_size + crops_coords_top_left + negative_target_size) - passed_add_embed_dim = ( - self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 - ) + passed_add_embed_dim = self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features - if expected_add_embed_dim != passed_add_embed_dim: + if ( + expected_add_embed_dim > passed_add_embed_dim + and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim + ): + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to enable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=True)` to make sure `aesthetic_score` {aesthetic_score} and `negative_aesthetic_score` {negative_aesthetic_score} is correctly used by the model." + ) + elif ( + expected_add_embed_dim < passed_add_embed_dim + and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim + ): raise ValueError( - f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`." + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to disable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=False)` to make sure `target_size` {target_size} is correctly used by the model." + ) + elif expected_add_embed_dim != passed_add_embed_dim: + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder.config.projection_dim`." ) add_time_ids = torch.tensor([add_time_ids], dtype=dtype) - return add_time_ids + add_neg_time_ids = torch.tensor([add_neg_time_ids], dtype=dtype) + + return add_time_ids, add_neg_time_ids # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae def upcast_vae(self): @@ -931,41 +968,46 @@ def num_timesteps(self): @torch.no_grad() @replace_example_docstring(EXAMPLE_DOC_STRING) def __call__( - self, - prompt: Union[str, List[str]] = None, - image: PipelineImageInput = None, - control_image: PipelineImageInput = None, - height: Optional[int] = None, - width: Optional[int] = None, - strength: float = 0.8, - num_inference_steps: int = 50, - guidance_scale: float = 5.0, - negative_prompt: Optional[Union[str, List[str]]] = None, - num_images_per_prompt: Optional[int] = 1, - eta: float = 0.0, - guess_mode: bool = False, - generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, - latents: Optional[torch.Tensor] = None, - prompt_embeds: Optional[torch.Tensor] = None, - negative_prompt_embeds: Optional[torch.Tensor] = None, - pooled_prompt_embeds: Optional[torch.Tensor] = None, - negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, - ip_adapter_image: Optional[PipelineImageInput] = None, - ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, - output_type: Optional[str] = "pil", - return_dict: bool = True, - cross_attention_kwargs: Optional[Dict[str, Any]] = None, - controlnet_conditioning_scale: Union[float, List[float]] = 0.8, - control_guidance_start: Union[float, List[float]] = 0.0, - control_guidance_end: Union[float, List[float]] = 1.0, - original_size: Tuple[int, int] = None, - crops_coords_top_left: Tuple[int, int] = (0, 0), - target_size: Tuple[int, int] = None, - callback_on_step_end: Optional[ - Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] - ] = None, - callback_on_step_end_tensor_inputs: List[str] = ["latents"], - **kwargs, + self, + prompt: Union[str, List[str]] = None, + image: PipelineImageInput = None, + control_image: PipelineImageInput = None, + height: Optional[int] = None, + width: Optional[int] = None, + strength: float = 0.8, + num_inference_steps: int = 50, + guidance_scale: float = 5.0, + negative_prompt: Optional[Union[str, List[str]]] = None, + num_images_per_prompt: Optional[int] = 1, + eta: float = 0.0, + guess_mode: bool = False, + generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, + latents: Optional[torch.Tensor] = None, + prompt_embeds: Optional[torch.Tensor] = None, + negative_prompt_embeds: Optional[torch.Tensor] = None, + pooled_prompt_embeds: Optional[torch.Tensor] = None, + negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, + ip_adapter_image: Optional[PipelineImageInput] = None, + ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, + output_type: Optional[str] = "pil", + return_dict: bool = True, + cross_attention_kwargs: Optional[Dict[str, Any]] = None, + controlnet_conditioning_scale: Union[float, List[float]] = 0.8, + control_guidance_start: Union[float, List[float]] = 0.0, + control_guidance_end: Union[float, List[float]] = 1.0, + original_size: Tuple[int, int] = None, + crops_coords_top_left: Tuple[int, int] = (0, 0), + target_size: Tuple[int, int] = None, + negative_original_size: Optional[Tuple[int, int]] = None, + negative_crops_coords_top_left: Tuple[int, int] = (0, 0), + negative_target_size: Optional[Tuple[int, int]] = None, + aesthetic_score: float = 6.0, + negative_aesthetic_score: float = 2.5, + callback_on_step_end: Optional[ + Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] + ] = None, + callback_on_step_end_tensor_inputs: List[str] = ["latents"], + **kwargs, ): r""" Function invoked when calling the pipeline for generation. @@ -1077,6 +1119,29 @@ def __call__( For most cases, `target_size` should be set to the desired height and width of the generated image. If not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + To negatively condition the generation process based on a specific image resolution. Part of SDXL's + micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): + To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's + micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): + To negatively condition the generation process based on a target image resolution. It should be as same + as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more + information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. + aesthetic_score (`float`, *optional*, defaults to 6.0): + Used to simulate an aesthetic score of the generated image by influencing the positive text condition. + Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). + negative_aesthetic_score (`float`, *optional*, defaults to 2.5): + Part of SDXL's micro-conditioning as explained in section 2.2 of + [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to + simulate an aesthetic score of the generated image by influencing the negative text condition. callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*): A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of each denoising step during the inference. with the following arguments: `callback_on_step_end(self: @@ -1289,19 +1354,37 @@ def __call__( target_size = target_size or (height, width) # 7. Prepare added time ids & embeddings + if negative_original_size is None: + negative_original_size = original_size + if negative_target_size is None: + negative_target_size = target_size + add_text_embeds = pooled_prompt_embeds - add_time_ids = self._get_add_time_ids( - original_size, crops_coords_top_left, target_size, dtype=prompt_embeds.dtype + text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1]) + + add_time_ids, add_neg_time_ids = self._get_add_time_ids( + original_size, + crops_coords_top_left, + target_size, + aesthetic_score, + negative_aesthetic_score, + negative_original_size, + negative_crops_coords_top_left, + negative_target_size, + dtype=prompt_embeds.dtype, + text_encoder_projection_dim=text_encoder_projection_dim, ) if self.do_classifier_free_guidance: prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0) add_time_ids = torch.cat([add_time_ids, add_time_ids], dim=0) + add_neg_time_ids = torch.cat([add_neg_time_ids, add_neg_time_ids], dim=0) prompt_embeds = prompt_embeds.to(device) add_text_embeds = add_text_embeds.to(device) add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1) + add_neg_time_ids = add_neg_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1) # patch diffusers controlnet instance forward, undo # after denoising loop @@ -1316,10 +1399,12 @@ def __call__( cn_og_forward = cn_model.forward def _cn_patch_forward(*args, **kwargs): - encoder_hidden_states = kwargs['encoder_hidden_states'] + encoder_hidden_states = kwargs["encoder_hidden_states"] if cn_model.encoder_hid_proj is not None and cn_model.config.encoder_hid_dim_type == "text_proj": - encoder_hidden_states = cn_model.encoder_hid_proj(kwargs['encoder_hidden_states']) - kwargs.pop('encoder_hidden_states') + # Ensure encoder_hidden_states is on the same device as the projection layer + encoder_hidden_states = encoder_hidden_states.to(cn_model.encoder_hid_proj.weight.device) + encoder_hidden_states = cn_model.encoder_hid_proj(encoder_hidden_states) + kwargs.pop("encoder_hidden_states") return cn_og_forward(*args, encoder_hidden_states=encoder_hidden_states, **kwargs) cn_model.forward = _cn_patch_forward @@ -1335,7 +1420,11 @@ def _cn_patch_forward(*args, **kwargs): latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) - added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids} + added_cond_kwargs = { + "text_embeds": add_text_embeds, + "time_ids": add_time_ids, + "neg_time_ids": add_neg_time_ids, + } # controlnet(s) inference if guess_mode and self.do_classifier_free_guidance: @@ -1346,6 +1435,7 @@ def _cn_patch_forward(*args, **kwargs): controlnet_added_cond_kwargs = { "text_embeds": add_text_embeds.chunk(2)[1], "time_ids": add_time_ids.chunk(2)[1], + "neg_time_ids": add_neg_time_ids.chunk(2)[1], } else: control_model_input = latent_model_input @@ -1376,7 +1466,9 @@ def _cn_patch_forward(*args, **kwargs): # To apply the output of ControlNet to both the unconditional and conditional batches, # add 0 to the unconditional batch to keep it unchanged. down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples] - mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample]) + mid_block_res_sample = torch.cat( + [torch.zeros_like(mid_block_res_sample), mid_block_res_sample] + ) if ip_adapter_image is not None or ip_adapter_image_embeds is not None: added_cond_kwargs["image_embeds"] = image_embeds @@ -1401,6 +1493,23 @@ def _cn_patch_forward(*args, **kwargs): # compute the previous noisy sample x_t -> x_t-1 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] + if callback_on_step_end is not None: + callback_kwargs = {} + for k in callback_on_step_end_tensor_inputs: + callback_kwargs[k] = locals()[k] + callback_outputs = callback_on_step_end(self, i, t, callback_kwargs) + + latents = callback_outputs.pop("latents", latents) + prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) + negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds) + add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds) + negative_pooled_prompt_embeds = callback_outputs.pop( + "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds + ) + add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids) + add_neg_time_ids = callback_outputs.pop("add_neg_time_ids", add_neg_time_ids) + control_image = callback_outputs.pop("control_image", control_image) + # call the callback, if provided if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): progress_bar.update() diff --git a/examples/community/pipeline_controlnet_xl_kolors_inpaint.py b/examples/community/pipeline_controlnet_xl_kolors_inpaint.py index b673bf5cbb5e..68d1153d0dea 100644 --- a/examples/community/pipeline_controlnet_xl_kolors_inpaint.py +++ b/examples/community/pipeline_controlnet_xl_kolors_inpaint.py @@ -92,10 +92,16 @@ >>> control_image = make_canny_condition(init_image) >>> controlnet = ControlNetModel.from_pretrained( - ... "Kwai-Kolors/Kolors-ControlNet-Canny", torch_dtype=torch.float16 + ... "Kwai-Kolors/Kolors-ControlNet-Canny", + ... use_safetensors=True, + ... torch_dtype=torch.float16 ... ) >>> pipe = KolorsControlNetInpaintPipeline.from_pretrained( - ... "Kwai-Kolors/Kolors-diffusers", controlnet=controlnet, torch_dtype=torch.float16 + ... "Kwai-Kolors/Kolors-diffusers", + ... controlnet=controlnet, + ... variant="fp16", + ... use_safetensors=True, + ... torch_dtype=torch.float16 ... ) >>> pipe.enable_model_cpu_offload() @@ -116,7 +122,7 @@ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents def retrieve_latents( - encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" + encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" ): if hasattr(encoder_output, "latent_dist") and sample_mode == "sample": return encoder_output.latent_dist.sample(generator) @@ -130,12 +136,12 @@ def retrieve_latents( # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps def retrieve_timesteps( - scheduler, - num_inference_steps: Optional[int] = None, - device: Optional[Union[str, torch.device]] = None, - timesteps: Optional[List[int]] = None, - sigmas: Optional[List[float]] = None, - **kwargs, + scheduler, + num_inference_steps: Optional[int] = None, + device: Optional[Union[str, torch.device]] = None, + timesteps: Optional[List[int]] = None, + sigmas: Optional[List[float]] = None, + **kwargs, ): """ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles @@ -249,21 +255,24 @@ class KolorsControlNetInpaintPipeline( "add_time_ids", "negative_pooled_prompt_embeds", "add_neg_time_ids", + "mask", + "masked_image_latents", + "control_image", ] def __init__( - self, - vae: AutoencoderKL, - text_encoder: ChatGLMModel, - tokenizer: ChatGLMTokenizer, - unet: UNet2DConditionModel, - controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel], - scheduler: KarrasDiffusionSchedulers, - requires_aesthetics_score: bool = False, - force_zeros_for_empty_prompt: bool = True, - feature_extractor: CLIPImageProcessor = None, - image_encoder: CLIPVisionModelWithProjection = None, - add_watermarker: Optional[bool] = None, + self, + vae: AutoencoderKL, + text_encoder: ChatGLMModel, + tokenizer: ChatGLMTokenizer, + unet: UNet2DConditionModel, + controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel], + scheduler: KarrasDiffusionSchedulers, + requires_aesthetics_score: bool = False, + force_zeros_for_empty_prompt: bool = True, + feature_extractor: CLIPImageProcessor = None, + image_encoder: CLIPVisionModelWithProjection = None, + add_watermarker: Optional[bool] = None, ): super().__init__() @@ -298,17 +307,17 @@ def __init__( self.register_to_config(requires_aesthetics_score=requires_aesthetics_score) def encode_prompt( - self, - prompt, - device: Optional[torch.device] = None, - num_images_per_prompt: int = 1, - do_classifier_free_guidance: bool = True, - negative_prompt=None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - lora_scale: Optional[float] = None, + self, + prompt, + device: Optional[torch.device] = None, + num_images_per_prompt: int = 1, + do_classifier_free_guidance: bool = True, + negative_prompt=None, + prompt_embeds: Optional[torch.FloatTensor] = None, + negative_prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + lora_scale: Optional[float] = None, ): r""" Encodes the prompt into text encoder hidden states. @@ -376,10 +385,11 @@ def encode_prompt( return_tensors="pt", ).to(self._execution_device) output = text_encoder( - input_ids=text_inputs['input_ids'], - attention_mask=text_inputs['attention_mask'], - position_ids=text_inputs['position_ids'], - output_hidden_states=True) + input_ids=text_inputs["input_ids"], + attention_mask=text_inputs["attention_mask"], + position_ids=text_inputs["position_ids"], + output_hidden_states=True, + ) prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] bs_embed, seq_len, _ = prompt_embeds.shape @@ -431,10 +441,11 @@ def encode_prompt( return_tensors="pt", ).to(self._execution_device) output = text_encoder( - input_ids=uncond_input['input_ids'], - attention_mask=uncond_input['attention_mask'], - position_ids=uncond_input['position_ids'], - output_hidden_states=True) + input_ids=uncond_input["input_ids"], + attention_mask=uncond_input["attention_mask"], + position_ids=uncond_input["position_ids"], + output_hidden_states=True, + ) negative_prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() negative_pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] @@ -496,7 +507,7 @@ def encode_image(self, image, device, num_images_per_prompt, output_hidden_state # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds def prepare_ip_adapter_image_embeds( - self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance + self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance ): image_embeds = [] if do_classifier_free_guidance: @@ -512,7 +523,7 @@ def prepare_ip_adapter_image_embeds( ) for single_ip_adapter_image, image_proj_layer in zip( - ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers + ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers ): output_hidden_state = not isinstance(image_proj_layer, ImageProjection) single_image_embeds, single_negative_image_embeds = self.encode_image( @@ -560,23 +571,23 @@ def prepare_extra_step_kwargs(self, generator, eta): return extra_step_kwargs def check_inputs( - self, - prompt, - image, - strength, - num_inference_steps, - callback_steps, - negative_prompt=None, - prompt_embeds=None, - negative_prompt_embeds=None, - pooled_prompt_embeds=None, - negative_pooled_prompt_embeds=None, - ip_adapter_image=None, - ip_adapter_image_embeds=None, - controlnet_conditioning_scale=1.0, - control_guidance_start=0.0, - control_guidance_end=1.0, - callback_on_step_end_tensor_inputs=None, + self, + prompt, + image, + strength, + num_inference_steps, + callback_steps, + negative_prompt=None, + prompt_embeds=None, + negative_prompt_embeds=None, + pooled_prompt_embeds=None, + negative_pooled_prompt_embeds=None, + ip_adapter_image=None, + ip_adapter_image_embeds=None, + controlnet_conditioning_scale=1.0, + control_guidance_start=0.0, + control_guidance_end=1.0, + callback_on_step_end_tensor_inputs=None, ): if strength < 0 or strength > 1: raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}") @@ -595,7 +606,7 @@ def check_inputs( ) if callback_on_step_end_tensor_inputs is not None and not all( - k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs + k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs ): raise ValueError( f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" @@ -652,15 +663,15 @@ def check_inputs( ) if ( - isinstance(self.controlnet, ControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, ControlNetModel) + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) ): self.check_image(image, prompt, prompt_embeds) elif ( - isinstance(self.controlnet, MultiControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) ): if not isinstance(image, list): raise TypeError("For multiple controlnets: `image` must be type `list`") @@ -681,22 +692,22 @@ def check_inputs( # Check `controlnet_conditioning_scale` if ( - isinstance(self.controlnet, ControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, ControlNetModel) + isinstance(self.controlnet, ControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, ControlNetModel) ): if not isinstance(controlnet_conditioning_scale, float): raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.") elif ( - isinstance(self.controlnet, MultiControlNetModel) - or is_compiled - and isinstance(self.controlnet._orig_mod, MultiControlNetModel) + isinstance(self.controlnet, MultiControlNetModel) + or is_compiled + and isinstance(self.controlnet._orig_mod, MultiControlNetModel) ): if isinstance(controlnet_conditioning_scale, list): if any(isinstance(i, list) for i in controlnet_conditioning_scale): raise ValueError("A single batch of multiple conditionings are supported at the moment.") elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len( - self.controlnet.nets + self.controlnet.nets ): raise ValueError( "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have" @@ -757,12 +768,12 @@ def check_image(self, image, prompt, prompt_embeds): image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray) if ( - not image_is_pil - and not image_is_tensor - and not image_is_np - and not image_is_pil_list - and not image_is_tensor_list - and not image_is_np_list + not image_is_pil + and not image_is_tensor + and not image_is_np + and not image_is_pil_list + and not image_is_tensor_list + and not image_is_np_list ): raise TypeError( f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}" @@ -787,16 +798,16 @@ def check_image(self, image, prompt, prompt_embeds): # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.prepare_image def prepare_control_image( - self, - image, - width, - height, - batch_size, - num_images_per_prompt, - device, - dtype, - do_classifier_free_guidance=False, - guess_mode=False, + self, + image, + width, + height, + batch_size, + num_images_per_prompt, + device, + dtype, + do_classifier_free_guidance=False, + guess_mode=False, ): image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) image_batch_size = image.shape[0] @@ -825,7 +836,7 @@ def get_timesteps(self, num_inference_steps, strength, device, denoising_start=N else: t_start = 0 - timesteps = self.scheduler.timesteps[t_start * self.scheduler.order:] + timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :] # Strength is irrelevant if we directly request a timestep to start at; # that is, strength is determined by the denoising_start instead. @@ -855,7 +866,7 @@ def get_timesteps(self, num_inference_steps, strength, device, denoising_start=N # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.prepare_latents def prepare_latents( - self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True + self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True ): if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)): raise ValueError( @@ -888,7 +899,7 @@ def prepare_latents( elif isinstance(generator, list): init_latents = [ - retrieve_latents(self.vae.encode(image[i: i + 1]), generator=generator[i]) + retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i]) for i in range(batch_size) ] init_latents = torch.cat(init_latents, dim=0) @@ -924,8 +935,9 @@ def prepare_latents( return latents # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents - def prepare_latents_t2i(self, batch_size, num_channels_latents, height, width, dtype, device, generator, - latents=None): + def prepare_latents_t2i( + self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None + ): shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) if isinstance(generator, list) and len(generator) != batch_size: raise ValueError( @@ -942,23 +954,55 @@ def prepare_latents_t2i(self, batch_size, num_channels_latents, height, width, d latents = latents * self.scheduler.init_noise_sigma return latents - def _get_add_time_ids(self, original_size, crops_coords_top_left, target_size, dtype): - add_time_ids = list(original_size + crops_coords_top_left + target_size) + # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline._get_add_time_ids + def _get_add_time_ids( + self, + original_size, + crops_coords_top_left, + target_size, + aesthetic_score, + negative_aesthetic_score, + negative_original_size, + negative_crops_coords_top_left, + negative_target_size, + dtype, + text_encoder_projection_dim=None, + ): + if self.config.requires_aesthetics_score: + add_time_ids = list(original_size + crops_coords_top_left + (aesthetic_score,)) + add_neg_time_ids = list( + negative_original_size + negative_crops_coords_top_left + (negative_aesthetic_score,) + ) + else: + add_time_ids = list(original_size + crops_coords_top_left + target_size) + add_neg_time_ids = list(negative_original_size + crops_coords_top_left + negative_target_size) - passed_add_embed_dim = ( - self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 - ) + passed_add_embed_dim = self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features - if expected_add_embed_dim != passed_add_embed_dim: + if ( + expected_add_embed_dim > passed_add_embed_dim + and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim + ): + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to enable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=True)` to make sure `aesthetic_score` {aesthetic_score} and `negative_aesthetic_score` {negative_aesthetic_score} is correctly used by the model." + ) + elif ( + expected_add_embed_dim < passed_add_embed_dim + and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim + ): raise ValueError( - f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector " - f"of {passed_add_embed_dim} was created. The model has an incorrect config. Please check " - f"`unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`." + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to disable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=False)` to make sure `target_size` {target_size} is correctly used by the model." + ) + elif expected_add_embed_dim != passed_add_embed_dim: + raise ValueError( + f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder.config.projection_dim`." ) add_time_ids = torch.tensor([add_time_ids], dtype=dtype) - return add_time_ids + add_neg_time_ids = torch.tensor([add_neg_time_ids], dtype=dtype) + + return add_time_ids, add_neg_time_ids # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae def upcast_vae(self): @@ -1013,7 +1057,7 @@ def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator): if isinstance(generator, list): image_latents = [ - retrieve_latents(self.vae.encode(image[i: i + 1]), generator=generator[i]) + retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i]) for i in range(image.shape[0]) ] image_latents = torch.cat(image_latents, dim=0) @@ -1029,7 +1073,7 @@ def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator): return image_latents def prepare_mask_latents( - self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance + self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance ): # resize the mask to latents shape as we concatenate the mask to the latents # we do that before converting to dtype to avoid breaking in case we're using cpu_offload @@ -1084,54 +1128,54 @@ def prepare_mask_latents( @torch.no_grad() @replace_example_docstring(EXAMPLE_DOC_STRING) def __call__( - self, - prompt: Union[str, List[str]] = None, - image: PipelineImageInput = None, - mask_image: PipelineImageInput = None, - control_image: PipelineImageInput = None, - masked_image_latents: torch.Tensor = None, - height: Optional[int] = None, - width: Optional[int] = None, - padding_mask_crop: Optional[int] = None, - strength: float = 0.9999, - num_inference_steps: int = 50, - timesteps: List[int] = None, - sigmas: List[float] = None, - denoising_start: Optional[float] = None, - denoising_end: Optional[float] = None, - guidance_scale: float = 7.5, - negative_prompt: Optional[Union[str, List[str]]] = None, - num_images_per_prompt: Optional[int] = 1, - eta: float = 0.0, - guess_mode: bool = False, - generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, - latents: Optional[torch.Tensor] = None, - prompt_embeds: Optional[torch.Tensor] = None, - negative_prompt_embeds: Optional[torch.Tensor] = None, - pooled_prompt_embeds: Optional[torch.Tensor] = None, - negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, - ip_adapter_image: Optional[PipelineImageInput] = None, - ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, - output_type: Optional[str] = "pil", - return_dict: bool = True, - cross_attention_kwargs: Optional[Dict[str, Any]] = None, - controlnet_conditioning_scale: Union[float, List[float]] = 0.8, - control_guidance_start: Union[float, List[float]] = 0.0, - control_guidance_end: Union[float, List[float]] = 1.0, - guidance_rescale: float = 0.0, - original_size: Tuple[int, int] = None, - crops_coords_top_left: Tuple[int, int] = (0, 0), - target_size: Tuple[int, int] = None, - negative_original_size: Optional[Tuple[int, int]] = None, - negative_crops_coords_top_left: Tuple[int, int] = (0, 0), - negative_target_size: Optional[Tuple[int, int]] = None, - aesthetic_score: float = 6.0, - negative_aesthetic_score: float = 2.5, - callback_on_step_end: Optional[ - Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] - ] = None, - callback_on_step_end_tensor_inputs: List[str] = ["latents"], - **kwargs, + self, + prompt: Union[str, List[str]] = None, + image: PipelineImageInput = None, + mask_image: PipelineImageInput = None, + control_image: PipelineImageInput = None, + masked_image_latents: torch.Tensor = None, + height: Optional[int] = None, + width: Optional[int] = None, + padding_mask_crop: Optional[int] = None, + strength: float = 0.9999, + num_inference_steps: int = 50, + timesteps: List[int] = None, + sigmas: List[float] = None, + denoising_start: Optional[float] = None, + denoising_end: Optional[float] = None, + guidance_scale: float = 7.5, + negative_prompt: Optional[Union[str, List[str]]] = None, + num_images_per_prompt: Optional[int] = 1, + eta: float = 0.0, + guess_mode: bool = False, + generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, + latents: Optional[torch.Tensor] = None, + prompt_embeds: Optional[torch.Tensor] = None, + negative_prompt_embeds: Optional[torch.Tensor] = None, + pooled_prompt_embeds: Optional[torch.Tensor] = None, + negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, + ip_adapter_image: Optional[PipelineImageInput] = None, + ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, + output_type: Optional[str] = "pil", + return_dict: bool = True, + cross_attention_kwargs: Optional[Dict[str, Any]] = None, + controlnet_conditioning_scale: Union[float, List[float]] = 0.8, + control_guidance_start: Union[float, List[float]] = 0.0, + control_guidance_end: Union[float, List[float]] = 1.0, + guidance_rescale: float = 0.0, + original_size: Tuple[int, int] = None, + crops_coords_top_left: Tuple[int, int] = (0, 0), + target_size: Tuple[int, int] = None, + negative_original_size: Optional[Tuple[int, int]] = None, + negative_crops_coords_top_left: Tuple[int, int] = (0, 0), + negative_target_size: Optional[Tuple[int, int]] = None, + aesthetic_score: float = 6.0, + negative_aesthetic_score: float = 2.5, + callback_on_step_end: Optional[ + Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] + ] = None, + callback_on_step_end_tensor_inputs: List[str] = ["latents"], + **kwargs, ): r""" Function invoked when calling the pipeline for generation. @@ -1598,29 +1642,47 @@ def denoising_value_valid(dnv): original_size = original_size or control_image.shape[-2:] target_size = target_size or (height, width) + if negative_original_size is None: + negative_original_size = original_size + if negative_target_size is None: + negative_target_size = target_size + add_text_embeds = pooled_prompt_embeds - add_time_ids = self._get_add_time_ids( - original_size, crops_coords_top_left, target_size, dtype=prompt_embeds.dtype + text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1]) + + add_time_ids, add_neg_time_ids = self._get_add_time_ids( + original_size, + crops_coords_top_left, + target_size, + aesthetic_score, + negative_aesthetic_score, + negative_original_size, + negative_crops_coords_top_left, + negative_target_size, + dtype=prompt_embeds.dtype, + text_encoder_projection_dim=text_encoder_projection_dim, ) if self.do_classifier_free_guidance: prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0) add_time_ids = torch.cat([add_time_ids, add_time_ids], dim=0) + add_neg_time_ids = torch.cat([add_neg_time_ids, add_neg_time_ids], dim=0) prompt_embeds = prompt_embeds.to(device) add_text_embeds = add_text_embeds.to(device) add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1) + add_neg_time_ids = add_neg_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1) # 10. Denoising loop num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0) if ( - self.denoising_end is not None - and self.denoising_start is not None - and denoising_value_valid(self.denoising_end) - and denoising_value_valid(self.denoising_start) - and self.denoising_start >= self.denoising_end + self.denoising_end is not None + and self.denoising_start is not None + and denoising_value_valid(self.denoising_end) + and denoising_value_valid(self.denoising_start) + and self.denoising_start >= self.denoising_end ): raise ValueError( f"`denoising_start`: {self.denoising_start} cannot be larger than or equal to `denoising_end`: " @@ -1657,10 +1719,12 @@ def denoising_value_valid(dnv): cn_og_forward = cn_model.forward def _cn_patch_forward(*args, **kwargs): - encoder_hidden_states = kwargs['encoder_hidden_states'] + encoder_hidden_states = kwargs["encoder_hidden_states"] if cn_model.encoder_hid_proj is not None and cn_model.config.encoder_hid_dim_type == "text_proj": - encoder_hidden_states = cn_model.encoder_hid_proj(kwargs['encoder_hidden_states']) - kwargs.pop('encoder_hidden_states') + # Ensure encoder_hidden_states is on the same device as the projection layer + encoder_hidden_states = encoder_hidden_states.to(cn_model.encoder_hid_proj.weight.device) + encoder_hidden_states = cn_model.encoder_hid_proj(encoder_hidden_states) + kwargs.pop("encoder_hidden_states") return cn_og_forward(*args, encoder_hidden_states=encoder_hidden_states, **kwargs) cn_model.forward = _cn_patch_forward @@ -1676,7 +1740,11 @@ def _cn_patch_forward(*args, **kwargs): if num_channels_unet == 9: latent_model_input = torch.cat([latent_model_input, mask, masked_image_latents], dim=1) - added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids} + added_cond_kwargs = { + "text_embeds": add_text_embeds, + "time_ids": add_time_ids, + "neg_time_ids": add_neg_time_ids, + } # controlnet(s) inference if guess_mode and self.do_classifier_free_guidance: @@ -1687,6 +1755,7 @@ def _cn_patch_forward(*args, **kwargs): controlnet_added_cond_kwargs = { "text_embeds": add_text_embeds.chunk(2)[1], "time_ids": add_time_ids.chunk(2)[1], + "neg_time_ids": add_neg_time_ids.chunk(2)[1], } else: control_model_input = latent_model_input @@ -1717,7 +1786,9 @@ def _cn_patch_forward(*args, **kwargs): # To apply the output of ControlNet to both the unconditional and conditional batches, # add 0 to the unconditional batch to keep it unchanged. down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples] - mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample]) + mid_block_res_sample = torch.cat( + [torch.zeros_like(mid_block_res_sample), mid_block_res_sample] + ) if ip_adapter_image is not None or ip_adapter_image_embeds is not None: added_cond_kwargs["image_embeds"] = image_embeds @@ -1751,14 +1822,7 @@ def _cn_patch_forward(*args, **kwargs): latents = callback_outputs.pop("latents", latents) prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds) - add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds) - negative_pooled_prompt_embeds = callback_outputs.pop( - "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds - ) - add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids) - add_neg_time_ids = callback_outputs.pop("add_neg_time_ids", add_neg_time_ids) - mask = callback_outputs.pop("mask", mask) - masked_image_latents = callback_outputs.pop("masked_image_latents", masked_image_latents) + control_image = callback_outputs.pop("control_image", control_image) # call the callback, if provided if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): diff --git a/examples/community/pipeline_kolors_inpainting.py b/examples/community/pipeline_kolors_inpainting.py index d6b83c0a1825..43a3957c82fe 100644 --- a/examples/community/pipeline_kolors_inpainting.py +++ b/examples/community/pipeline_kolors_inpainting.py @@ -76,8 +76,9 @@ ... "Kwai-Kolors/Kolors-diffusers", ... torch_dtype=torch.float16, ... variant="fp16" + ... use_safetensors=True ... ) - >>> pipe.to("cuda") + >>> pipe.enable_model_cpu_offload() >>> img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png" >>> mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png" @@ -244,7 +245,7 @@ def prepare_mask_and_masked_image(image, mask, height, width, return_image: bool # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents def retrieve_latents( - encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" + encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" ): if hasattr(encoder_output, "latent_dist") and sample_mode == "sample": return encoder_output.latent_dist.sample(generator) @@ -258,12 +259,12 @@ def retrieve_latents( # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps def retrieve_timesteps( - scheduler, - num_inference_steps: Optional[int] = None, - device: Optional[Union[str, torch.device]] = None, - timesteps: Optional[List[int]] = None, - sigmas: Optional[List[float]] = None, - **kwargs, + scheduler, + num_inference_steps: Optional[int] = None, + device: Optional[Union[str, torch.device]] = None, + timesteps: Optional[List[int]] = None, + sigmas: Optional[List[float]] = None, + **kwargs, ): """ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles @@ -379,17 +380,17 @@ class KolorsInpaintPipeline( ] def __init__( - self, - vae: AutoencoderKL, - text_encoder: ChatGLMModel, - tokenizer: ChatGLMTokenizer, - unet: UNet2DConditionModel, - scheduler: KarrasDiffusionSchedulers, - image_encoder: CLIPVisionModelWithProjection = None, - feature_extractor: CLIPImageProcessor = None, - requires_aesthetics_score: bool = False, - force_zeros_for_empty_prompt: bool = True, - add_watermarker: Optional[bool] = None, + self, + vae: AutoencoderKL, + text_encoder: ChatGLMModel, + tokenizer: ChatGLMTokenizer, + unet: UNet2DConditionModel, + scheduler: KarrasDiffusionSchedulers, + image_encoder: CLIPVisionModelWithProjection = None, + feature_extractor: CLIPImageProcessor = None, + requires_aesthetics_score: bool = False, + force_zeros_for_empty_prompt: bool = True, + add_watermarker: Optional[bool] = None, ): super().__init__() @@ -444,7 +445,7 @@ def encode_image(self, image, device, num_images_per_prompt, output_hidden_state # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds def prepare_ip_adapter_image_embeds( - self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance + self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance ): if ip_adapter_image_embeds is None: if not isinstance(ip_adapter_image, list): @@ -457,7 +458,7 @@ def prepare_ip_adapter_image_embeds( image_embeds = [] for single_ip_adapter_image, image_proj_layer in zip( - ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers + ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers ): output_hidden_state = not isinstance(image_proj_layer, ImageProjection) single_image_embeds, single_negative_image_embeds = self.encode_image( @@ -495,17 +496,17 @@ def prepare_ip_adapter_image_embeds( return image_embeds def encode_prompt( - self, - prompt, - device: Optional[torch.device] = None, - num_images_per_prompt: int = 1, - do_classifier_free_guidance: bool = True, - negative_prompt=None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - lora_scale: Optional[float] = None, + self, + prompt, + device: Optional[torch.device] = None, + num_images_per_prompt: int = 1, + do_classifier_free_guidance: bool = True, + negative_prompt=None, + prompt_embeds: Optional[torch.FloatTensor] = None, + negative_prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + lora_scale: Optional[float] = None, ): r""" Encodes the prompt into text encoder hidden states. @@ -573,10 +574,11 @@ def encode_prompt( return_tensors="pt", ).to(self._execution_device) output = text_encoder( - input_ids=text_inputs['input_ids'], - attention_mask=text_inputs['attention_mask'], - position_ids=text_inputs['position_ids'], - output_hidden_states=True) + input_ids=text_inputs["input_ids"], + attention_mask=text_inputs["attention_mask"], + position_ids=text_inputs["position_ids"], + output_hidden_states=True, + ) prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] bs_embed, seq_len, _ = prompt_embeds.shape @@ -628,10 +630,11 @@ def encode_prompt( return_tensors="pt", ).to(self._execution_device) output = text_encoder( - input_ids=uncond_input['input_ids'], - attention_mask=uncond_input['attention_mask'], - position_ids=uncond_input['position_ids'], - output_hidden_states=True) + input_ids=uncond_input["input_ids"], + attention_mask=uncond_input["attention_mask"], + position_ids=uncond_input["position_ids"], + output_hidden_states=True, + ) negative_prompt_embeds = output.hidden_states[-2].permute(1, 0, 2).clone() negative_pooled_prompt_embeds = output.hidden_states[-1][-1, :, :].clone() # [batch_size, 4096] @@ -685,22 +688,22 @@ def prepare_extra_step_kwargs(self, generator, eta): return extra_step_kwargs def check_inputs( - self, - prompt, - image, - mask_image, - height, - width, - strength, - callback_steps, - output_type, - negative_prompt=None, - prompt_embeds=None, - negative_prompt_embeds=None, - ip_adapter_image=None, - ip_adapter_image_embeds=None, - callback_on_step_end_tensor_inputs=None, - padding_mask_crop=None, + self, + prompt, + image, + mask_image, + height, + width, + strength, + callback_steps, + output_type, + negative_prompt=None, + prompt_embeds=None, + negative_prompt_embeds=None, + ip_adapter_image=None, + ip_adapter_image_embeds=None, + callback_on_step_end_tensor_inputs=None, + padding_mask_crop=None, ): if strength < 0 or strength > 1: raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}") @@ -715,7 +718,7 @@ def check_inputs( ) if callback_on_step_end_tensor_inputs is not None and not all( - k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs + k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs ): raise ValueError( f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" @@ -749,7 +752,7 @@ def check_inputs( if padding_mask_crop is not None: if not isinstance(image, PIL.Image.Image): raise ValueError( - f"The image should be a PIL image when inpainting mask crop, but is of type" f" {type(image)}." + f"The image should be a PIL image when inpainting mask crop, but is of type {type(image)}." ) if not isinstance(mask_image, PIL.Image.Image): raise ValueError( @@ -757,7 +760,7 @@ def check_inputs( f" {type(mask_image)}." ) if output_type != "pil": - raise ValueError(f"The output type should be PIL when inpainting mask crop, but is" f" {output_type}.") + raise ValueError(f"The output type should be PIL when inpainting mask crop, but is {output_type}.") if ip_adapter_image is not None and ip_adapter_image_embeds is not None: raise ValueError( @@ -775,21 +778,21 @@ def check_inputs( ) def prepare_latents( - self, - batch_size, - num_channels_latents, - height, - width, - dtype, - device, - generator, - latents=None, - image=None, - timestep=None, - is_strength_max=True, - add_noise=True, - return_noise=False, - return_image_latents=False, + self, + batch_size, + num_channels_latents, + height, + width, + dtype, + device, + generator, + latents=None, + image=None, + timestep=None, + is_strength_max=True, + add_noise=True, + return_noise=False, + return_image_latents=False, ): shape = ( batch_size, @@ -848,7 +851,7 @@ def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator): if isinstance(generator, list): image_latents = [ - retrieve_latents(self.vae.encode(image[i: i + 1]), generator=generator[i]) + retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i]) for i in range(image.shape[0]) ] image_latents = torch.cat(image_latents, dim=0) @@ -864,7 +867,7 @@ def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator): return image_latents def prepare_mask_latents( - self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance + self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance ): # resize the mask to latents shape as we concatenate the mask to the latents # we do that before converting to dtype to avoid breaking in case we're using cpu_offload @@ -925,7 +928,7 @@ def get_timesteps(self, num_inference_steps, strength, device, denoising_start=N else: t_start = 0 - timesteps = self.scheduler.timesteps[t_start * self.scheduler.order:] + timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :] # Strength is irrelevant if we directly request a timestep to start at; # that is, strength is determined by the denoising_start instead. @@ -955,17 +958,17 @@ def get_timesteps(self, num_inference_steps, strength, device, denoising_start=N # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline._get_add_time_ids def _get_add_time_ids( - self, - original_size, - crops_coords_top_left, - target_size, - aesthetic_score, - negative_aesthetic_score, - negative_original_size, - negative_crops_coords_top_left, - negative_target_size, - dtype, - text_encoder_projection_dim=None, + self, + original_size, + crops_coords_top_left, + target_size, + aesthetic_score, + negative_aesthetic_score, + negative_original_size, + negative_crops_coords_top_left, + negative_target_size, + dtype, + text_encoder_projection_dim=None, ): if self.config.requires_aesthetics_score: add_time_ids = list(original_size + crops_coords_top_left + (aesthetic_score,)) @@ -976,21 +979,19 @@ def _get_add_time_ids( add_time_ids = list(original_size + crops_coords_top_left + target_size) add_neg_time_ids = list(negative_original_size + crops_coords_top_left + negative_target_size) - passed_add_embed_dim = ( - self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 - ) + passed_add_embed_dim = self.unet.config.addition_time_embed_dim * len(add_time_ids) + 4096 expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features if ( - expected_add_embed_dim > passed_add_embed_dim - and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim + expected_add_embed_dim > passed_add_embed_dim + and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim ): raise ValueError( f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to enable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=True)` to make sure `aesthetic_score` {aesthetic_score} and `negative_aesthetic_score` {negative_aesthetic_score} is correctly used by the model." ) elif ( - expected_add_embed_dim < passed_add_embed_dim - and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim + expected_add_embed_dim < passed_add_embed_dim + and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim ): raise ValueError( f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to disable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=False)` to make sure `target_size` {target_size} is correctly used by the model." @@ -1027,7 +1028,7 @@ def upcast_vae(self): # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding def get_guidance_scale_embedding( - self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32 + self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32 ) -> torch.Tensor: """ See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298 @@ -1094,49 +1095,49 @@ def interrupt(self): @torch.no_grad() @replace_example_docstring(EXAMPLE_DOC_STRING) def __call__( - self, - prompt: Union[str, List[str]] = None, - image: PipelineImageInput = None, - mask_image: PipelineImageInput = None, - masked_image_latents: torch.Tensor = None, - height: Optional[int] = None, - width: Optional[int] = None, - padding_mask_crop: Optional[int] = None, - strength: float = 0.9999, - num_inference_steps: int = 50, - timesteps: List[int] = None, - sigmas: List[float] = None, - denoising_start: Optional[float] = None, - denoising_end: Optional[float] = None, - guidance_scale: float = 7.5, - negative_prompt: Optional[Union[str, List[str]]] = None, - num_images_per_prompt: Optional[int] = 1, - eta: float = 0.0, - generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, - latents: Optional[torch.Tensor] = None, - prompt_embeds: Optional[torch.Tensor] = None, - negative_prompt_embeds: Optional[torch.Tensor] = None, - pooled_prompt_embeds: Optional[torch.Tensor] = None, - negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, - ip_adapter_image: Optional[PipelineImageInput] = None, - ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, - output_type: Optional[str] = "pil", - return_dict: bool = True, - cross_attention_kwargs: Optional[Dict[str, Any]] = None, - guidance_rescale: float = 0.0, - original_size: Tuple[int, int] = None, - crops_coords_top_left: Tuple[int, int] = (0, 0), - target_size: Tuple[int, int] = None, - negative_original_size: Optional[Tuple[int, int]] = None, - negative_crops_coords_top_left: Tuple[int, int] = (0, 0), - negative_target_size: Optional[Tuple[int, int]] = None, - aesthetic_score: float = 6.0, - negative_aesthetic_score: float = 2.5, - callback_on_step_end: Optional[ - Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] - ] = None, - callback_on_step_end_tensor_inputs: List[str] = ["latents"], - **kwargs, + self, + prompt: Union[str, List[str]] = None, + image: PipelineImageInput = None, + mask_image: PipelineImageInput = None, + masked_image_latents: torch.Tensor = None, + height: Optional[int] = None, + width: Optional[int] = None, + padding_mask_crop: Optional[int] = None, + strength: float = 0.9999, + num_inference_steps: int = 50, + timesteps: List[int] = None, + sigmas: List[float] = None, + denoising_start: Optional[float] = None, + denoising_end: Optional[float] = None, + guidance_scale: float = 7.5, + negative_prompt: Optional[Union[str, List[str]]] = None, + num_images_per_prompt: Optional[int] = 1, + eta: float = 0.0, + generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, + latents: Optional[torch.Tensor] = None, + prompt_embeds: Optional[torch.Tensor] = None, + negative_prompt_embeds: Optional[torch.Tensor] = None, + pooled_prompt_embeds: Optional[torch.Tensor] = None, + negative_pooled_prompt_embeds: Optional[torch.Tensor] = None, + ip_adapter_image: Optional[PipelineImageInput] = None, + ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None, + output_type: Optional[str] = "pil", + return_dict: bool = True, + cross_attention_kwargs: Optional[Dict[str, Any]] = None, + guidance_rescale: float = 0.0, + original_size: Tuple[int, int] = None, + crops_coords_top_left: Tuple[int, int] = (0, 0), + target_size: Tuple[int, int] = None, + negative_original_size: Optional[Tuple[int, int]] = None, + negative_crops_coords_top_left: Tuple[int, int] = (0, 0), + negative_target_size: Optional[Tuple[int, int]] = None, + aesthetic_score: float = 6.0, + negative_aesthetic_score: float = 2.5, + callback_on_step_end: Optional[ + Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] + ] = None, + callback_on_step_end_tensor_inputs: List[str] = ["latents"], + **kwargs, ): r""" Function invoked when calling the pipeline for generation. @@ -1558,11 +1559,11 @@ def denoising_value_valid(dnv): num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0) if ( - self.denoising_end is not None - and self.denoising_start is not None - and denoising_value_valid(self.denoising_end) - and denoising_value_valid(self.denoising_start) - and self.denoising_start >= self.denoising_end + self.denoising_end is not None + and self.denoising_start is not None + and denoising_value_valid(self.denoising_end) + and denoising_value_valid(self.denoising_start) + and self.denoising_start >= self.denoising_end ): raise ValueError( f"`denoising_start`: {self.denoising_start} cannot be larger than or equal to `denoising_end`: "