import torch import numpy as np from typing import Any, Dict, List, Optional, Tuple, Union from typing import Any, Callable, List from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback from diffusers.image_processor import PipelineImageInput from diffusers.pipelines.stable_diffusion_xl import StableDiffusionXLPipeline from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl import retrieve_timesteps from diffusers.utils import deprecate, is_torch_xla_available from src.ptp_sdxl.processors import P2PCrossAttnProcessor, create_controller from src.ptp_sdxl.ipa_p2p_processor import IPAP2PCrossAttnProcessor if is_torch_xla_available(): import torch_xla.core.xla_model as xm XLA_AVAILABLE = True else: XLA_AVAILABLE = False # 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 class IPAPrompt2PromptPipeline(StableDiffusionXLPipeline): r""" Args: Prompt-to-Prompt-Pipeline for text-to-image generation using Stable Diffusion. This model inherits from [`StableDiffusionPipeline`]. 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.) vae ([`AutoencoderKL`]): Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. text_encoder ([`CLIPTextModel`]): Frozen text-encoder. Stable Diffusion uses the text portion of [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant. tokenizer (`CLIPTokenizer`): Tokenizer of class [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer). 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`]. safety_checker ([`StableDiffusionSafetyChecker`]): Classification module that estimates whether generated images could be considered offensive or harmful. Please, refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for details. feature_extractor ([`CLIPFeatureExtractor`]): Model that extracts features from generated images to be used as inputs for the `safety_checker`. """ _optional_components = ["safety_checker", "feature_extractor"] def check_inputs( self, prompt, prompt_2, height, width, callback_steps, negative_prompt=None, negative_prompt_2=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, callback_on_step_end_tensor_inputs=None, ): 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_2 is not None and prompt_embeds is not None: raise ValueError( f"Cannot forward both `prompt_2`: {prompt_2} 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)}") elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)): raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}") 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." ) elif negative_prompt_2 is not None and negative_prompt_embeds is not None: raise ValueError( f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} 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`." ) 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 _aggregate_and_get_attention_maps_per_token(self, with_softmax): attention_maps = self.controller.aggregate_attention( from_where=("up_cross", "down_cross", "mid_cross"), # from_where=("up", "down"), # from_where=("down",) ) attention_maps_list = self._get_attention_maps_list( attention_maps=attention_maps, with_softmax=with_softmax ) return attention_maps_list @staticmethod def _get_attention_maps_list( attention_maps: torch.Tensor, with_softmax ) -> List[torch.Tensor]: attention_maps *= 100 if with_softmax: attention_maps = torch.nn.functional.softmax(attention_maps, dim=-1) attention_maps_list = [ attention_maps[:, :, i] for i in range(attention_maps.shape[2]) ] return attention_maps_list @torch.no_grad() def __call__( self, prompt: Union[str, List[str]] = None, prompt_2: Optional[Union[str, List[str]]] = None, height: Optional[int] = None, width: Optional[int] = None, num_inference_steps: int = 50, timesteps: List[int] = None, sigmas: List[float] = None, denoising_end: Optional[float] = None, guidance_scale: float = 5.0, negative_prompt: Optional[Union[str, List[str]]] = None, negative_prompt_2: 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: Optional[Tuple[int, int]] = None, crops_coords_top_left: Tuple[int, int] = (0, 0), target_size: Optional[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, clip_skip: Optional[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"], attn_res=None, **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 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. 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_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 as determined by the discrete timesteps selected by 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 5.0): 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 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_xl.StableDiffusionXLPipelineOutput`] 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). guidance_rescale (`float`, *optional*, defaults to 0.0): Guidance rescale factor proposed by [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf) `guidance_scale` is defined as `φ` in equation 16. of [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). Guidance rescale factor should fix overexposure when using zero terminal SNR. 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. 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_xl.StableDiffusionXLPipelineOutput`] or `tuple`: [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] if `return_dict` is True, otherwise a `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.default_sample_size * self.vae_scale_factor width = width or self.default_sample_size * self.vae_scale_factor original_size = original_size or (height, width) target_size = target_size or (height, width) if attn_res is None: attn_res = int(np.ceil(width / 32)), int(np.ceil(height / 32)) self.attn_res = attn_res self.controller = create_controller( prompt, cross_attention_kwargs, num_inference_steps, tokenizer=self.tokenizer, device=self.device, attn_res=self.attn_res ) self.register_attention_control_consider_ipa(self.controller) # add attention controller cross_attention_kwargs.pop("edit_type", None) cross_attention_kwargs.pop("n_self_replace", None) cross_attention_kwargs.pop("n_cross_replace", None) cross_attention_kwargs.pop("equalizer_words", None) cross_attention_kwargs.pop("equalizer_strengths", None) cross_attention_kwargs.pop("local_blend_words", None) # 1. Check inputs. Raise error if not correct self.check_inputs( prompt, prompt_2, height, width, callback_steps, negative_prompt, negative_prompt_2, prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds, ip_adapter_image, ip_adapter_image_embeds, callback_on_step_end_tensor_inputs, ) self._guidance_scale = guidance_scale self._guidance_rescale = guidance_rescale self._clip_skip = clip_skip self._cross_attention_kwargs = cross_attention_kwargs self._denoising_end = denoising_end 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 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, prompt_2=prompt_2, device=device, num_images_per_prompt=num_images_per_prompt, do_classifier_free_guidance=self.do_classifier_free_guidance, negative_prompt=negative_prompt, negative_prompt_2=negative_prompt_2, 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=lora_scale, clip_skip=self.clip_skip, ) # 4. Prepare timesteps timesteps, num_inference_steps = retrieve_timesteps( self.scheduler, num_inference_steps, device, timesteps, sigmas ) # 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, ) # latents[1] = latents[0] # maybe should set all batch not just the first one latents = torch.cat([latents] * 2, dim=0) # 6. 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. Prepare added time ids & embeddings add_text_embeds = pooled_prompt_embeds if self.text_encoder_2 is None: text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1]) else: text_encoder_projection_dim = self.text_encoder_2.config.projection_dim add_time_ids = self._get_add_time_ids( 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) add_time_ids = torch.cat([negative_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) 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, ) # 8. Denoising loop num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0) # 8.1 Apply denoising_end if ( self.denoising_end is not None and isinstance(self.denoising_end, float) and self.denoising_end > 0 and self.denoising_end < 1 ): 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] # 9. 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 latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) # 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) # step callback latents = self.controller.step_callback(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) negative_add_time_ids = callback_outputs.pop("negative_add_time_ids", negative_add_time_ids) # 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() last_latents = latents.clone() 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: image = latents if not output_type == "latent": # 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) # Offload all models self.maybe_free_model_hooks() if not return_dict: return (image,) out = StableDiffusionXLPipelineOutput(images=image) out.last_latents = last_latents return out def register_attention_control_consider_ipa(self, controller): attn_procs = {} cross_att_count = 0 for name in self.unet.attn_processors.keys(): # print(f"attention name: {name} and processor: {self.unet.attn_processors[name].__class__.__name__}") None if name.endswith("attn1.processor") else self.unet.config.cross_attention_dim if name.startswith("mid_block"): self.unet.config.block_out_channels[-1] place_in_unet = "mid" elif name.startswith("up_blocks"): block_id = int(name[len("up_blocks.")]) list(reversed(self.unet.config.block_out_channels))[block_id] place_in_unet = "up" elif name.startswith("down_blocks"): block_id = int(name[len("down_blocks.")]) self.unet.config.block_out_channels[block_id] place_in_unet = "down" else: attn_procs[name] = self.unet.attn_processors[name] continue cross_att_count += 1 if self.unet.attn_processors[name].__class__.__name__ in ["IPAdapterAttnProcessor", "IPAdapterAttnProcessor2_0", "IPAP2PCrossAttnProcessor"]: attn_procs[name] = IPAP2PCrossAttnProcessor( controller=controller, place_in_unet=place_in_unet, ipa_processor=self.unet.attn_processors[name] ) else: attn_procs[name] = P2PCrossAttnProcessor(controller=controller, place_in_unet=place_in_unet) self.unet.set_attn_processor(attn_procs) controller.num_att_layers = cross_att_count