# 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, Union import torch import torch.nn as nn import torch.nn.functional as F from packaging import version from safetensors import safe_open from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection from diffusers.configuration_utils import FrozenDict from diffusers.image_processor import VaeImageProcessor from diffusers.loaders import ( FromSingleFileMixin, IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin, ) from diffusers.models import AutoencoderKL, UNet2DConditionModel from diffusers.models.attention_processor import ( AttnProcessor, AttnProcessor2_0, IPAdapterAttnProcessor, IPAdapterAttnProcessor2_0, ) from diffusers.models.embeddings import MultiIPAdapterImageProjection from diffusers.models.lora import adjust_lora_scale_text_encoder from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin from diffusers.pipelines.stable_diffusion.pipeline_output import StableDiffusionPipelineOutput from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker from diffusers.schedulers import KarrasDiffusionSchedulers from diffusers.utils import ( USE_PEFT_BACKEND, _get_model_file, deprecate, logging, scale_lora_layers, unscale_lora_layers, ) from diffusers.utils.torch_utils import randn_tensor logger = logging.get_logger(__name__) # pylint: disable=invalid-name class IPAdapterFullImageProjection(nn.Module): def __init__(self, image_embed_dim=1024, cross_attention_dim=1024, mult=1, num_tokens=1): super().__init__() from diffusers.models.attention import FeedForward self.num_tokens = num_tokens self.cross_attention_dim = cross_attention_dim self.ff = FeedForward(image_embed_dim, cross_attention_dim * num_tokens, mult=mult, activation_fn="gelu") self.norm = nn.LayerNorm(cross_attention_dim) def forward(self, image_embeds: torch.Tensor): x = self.ff(image_embeds) x = x.reshape(-1, self.num_tokens, self.cross_attention_dim) return self.norm(x) 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 retrieve_timesteps( scheduler, num_inference_steps: Optional[int] = None, device: Optional[Union[str, torch.device]] = None, timesteps: Optional[List[int]] = 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 support arbitrary spacing between timesteps. If `None`, then the default timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps` 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: 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) else: scheduler.set_timesteps(num_inference_steps, device=device, **kwargs) timesteps = scheduler.timesteps return timesteps, num_inference_steps class IPAdapterFaceIDStableDiffusionPipeline( DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin, IPAdapterMixin, FromSingleFileMixin, ): r""" Pipeline for text-to-image generation using Stable Diffusion. This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular device, etc.). The pipeline also inherits the following loading methods: - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files - [`~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 ([`~transformers.CLIPTextModel`]): Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)). tokenizer ([`~transformers.CLIPTokenizer`]): A `CLIPTokenizer` to tokenize text. unet ([`UNet2DConditionModel`]): A `UNet2DConditionModel` 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 more details about a model's potential harms. 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 = ["safety_checker", "feature_extractor", "image_encoder"] _exclude_from_cpu_offload = ["safety_checker"] _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"] def __init__( self, vae: AutoencoderKL, text_encoder: CLIPTextModel, tokenizer: CLIPTokenizer, unet: UNet2DConditionModel, scheduler: KarrasDiffusionSchedulers, safety_checker: StableDiffusionSafetyChecker, feature_extractor: CLIPImageProcessor, image_encoder: CLIPVisionModelWithProjection = None, requires_safety_checker: bool = True, ): super().__init__() if hasattr(scheduler.config, "steps_offset") and scheduler.config.steps_offset != 1: deprecation_message = ( f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`" f" should be set to 1 instead of {scheduler.config.steps_offset}. Please make sure " "to update the config accordingly as leaving `steps_offset` might led to incorrect results" " in future versions. If you have downloaded this checkpoint from the Hugging Face Hub," " it would be very nice if you could open a Pull request for the `scheduler/scheduler_config.json`" " file" ) deprecate("steps_offset!=1", "1.0.0", deprecation_message, standard_warn=False) new_config = dict(scheduler.config) new_config["steps_offset"] = 1 scheduler._internal_dict = FrozenDict(new_config) if hasattr(scheduler.config, "clip_sample") and scheduler.config.clip_sample is True: deprecation_message = ( f"The configuration file of this scheduler: {scheduler} has not set the configuration `clip_sample`." " `clip_sample` should be set to False in the configuration file. Please make sure to update the" " config accordingly as not setting `clip_sample` in the config might lead to incorrect results in" " future versions. If you have downloaded this checkpoint from the Hugging Face Hub, it would be very" " nice if you could open a Pull request for the `scheduler/scheduler_config.json` file" ) deprecate("clip_sample not set", "1.0.0", deprecation_message, standard_warn=False) new_config = dict(scheduler.config) new_config["clip_sample"] = False scheduler._internal_dict = FrozenDict(new_config) if safety_checker is None and requires_safety_checker: logger.warning( f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure" " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered" " results in services or applications open to the public. Both the diffusers team and Hugging Face" " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling" " it only for use-cases that involve analyzing network behavior or auditing its results. For more" " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ." ) if safety_checker is not None and feature_extractor is None: raise ValueError( "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety" " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead." ) is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse( version.parse(unet.config._diffusers_version).base_version ) < version.parse("0.9.0.dev0") is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64 if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64: deprecation_message = ( "The configuration file of the unet has set the default `sample_size` to smaller than" " 64 which seems highly unlikely. If your checkpoint is a fine-tuned version of any of the" " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-" " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5" " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the" " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`" " in the config might lead to incorrect results in future versions. If you have downloaded this" " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for" " the `unet/config.json` file" ) deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False) new_config = dict(unet.config) new_config["sample_size"] = 64 unet._internal_dict = FrozenDict(new_config) self.register_modules( vae=vae, text_encoder=text_encoder, tokenizer=tokenizer, unet=unet, scheduler=scheduler, safety_checker=safety_checker, 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) self.register_to_config(requires_safety_checker=requires_safety_checker) def load_ip_adapter_face_id(self, pretrained_model_name_or_path_or_dict, weight_name, **kwargs): cache_dir = kwargs.pop("cache_dir", None) force_download = kwargs.pop("force_download", False) proxies = kwargs.pop("proxies", None) local_files_only = kwargs.pop("local_files_only", None) token = kwargs.pop("token", None) revision = kwargs.pop("revision", None) subfolder = kwargs.pop("subfolder", None) user_agent = { "file_type": "attn_procs_weights", "framework": "pytorch", } model_file = _get_model_file( pretrained_model_name_or_path_or_dict, weights_name=weight_name, cache_dir=cache_dir, force_download=force_download, proxies=proxies, local_files_only=local_files_only, token=token, revision=revision, subfolder=subfolder, user_agent=user_agent, ) if weight_name.endswith(".safetensors"): state_dict = {"image_proj": {}, "ip_adapter": {}} with safe_open(model_file, framework="pt", device="cpu") as f: for key in f.keys(): if key.startswith("image_proj."): state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key) elif key.startswith("ip_adapter."): state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key) else: state_dict = torch.load(model_file, map_location="cpu") self._load_ip_adapter_weights(state_dict) def convert_ip_adapter_image_proj_to_diffusers(self, state_dict): updated_state_dict = {} clip_embeddings_dim_in = state_dict["proj.0.weight"].shape[1] clip_embeddings_dim_out = state_dict["proj.0.weight"].shape[0] multiplier = clip_embeddings_dim_out // clip_embeddings_dim_in norm_layer = "norm.weight" cross_attention_dim = state_dict[norm_layer].shape[0] num_tokens = state_dict["proj.2.weight"].shape[0] // cross_attention_dim image_projection = IPAdapterFullImageProjection( cross_attention_dim=cross_attention_dim, image_embed_dim=clip_embeddings_dim_in, mult=multiplier, num_tokens=num_tokens, ) for key, value in state_dict.items(): diffusers_name = key.replace("proj.0", "ff.net.0.proj") diffusers_name = diffusers_name.replace("proj.2", "ff.net.2") updated_state_dict[diffusers_name] = value image_projection.load_state_dict(updated_state_dict) return image_projection def _load_ip_adapter_weights(self, state_dict): num_image_text_embeds = 4 self.unet.encoder_hid_proj = None # set ip-adapter cross-attention processors & load state_dict attn_procs = {} lora_dict = {} key_id = 0 for name in self.unet.attn_processors.keys(): cross_attention_dim = None if name.endswith("attn1.processor") else self.unet.config.cross_attention_dim if name.startswith("mid_block"): hidden_size = self.unet.config.block_out_channels[-1] elif name.startswith("up_blocks"): block_id = int(name[len("up_blocks.")]) hidden_size = list(reversed(self.unet.config.block_out_channels))[block_id] elif name.startswith("down_blocks"): block_id = int(name[len("down_blocks.")]) hidden_size = self.unet.config.block_out_channels[block_id] if cross_attention_dim is None or "motion_modules" in name: attn_processor_class = ( AttnProcessor2_0 if hasattr(F, "scaled_dot_product_attention") else AttnProcessor ) attn_procs[name] = attn_processor_class() lora_dict.update( {f"unet.{name}.to_k_lora.down.weight": state_dict["ip_adapter"][f"{key_id}.to_k_lora.down.weight"]} ) lora_dict.update( {f"unet.{name}.to_q_lora.down.weight": state_dict["ip_adapter"][f"{key_id}.to_q_lora.down.weight"]} ) lora_dict.update( {f"unet.{name}.to_v_lora.down.weight": state_dict["ip_adapter"][f"{key_id}.to_v_lora.down.weight"]} ) lora_dict.update( { f"unet.{name}.to_out_lora.down.weight": state_dict["ip_adapter"][ f"{key_id}.to_out_lora.down.weight" ] } ) lora_dict.update( {f"unet.{name}.to_k_lora.up.weight": state_dict["ip_adapter"][f"{key_id}.to_k_lora.up.weight"]} ) lora_dict.update( {f"unet.{name}.to_q_lora.up.weight": state_dict["ip_adapter"][f"{key_id}.to_q_lora.up.weight"]} ) lora_dict.update( {f"unet.{name}.to_v_lora.up.weight": state_dict["ip_adapter"][f"{key_id}.to_v_lora.up.weight"]} ) lora_dict.update( {f"unet.{name}.to_out_lora.up.weight": state_dict["ip_adapter"][f"{key_id}.to_out_lora.up.weight"]} ) key_id += 1 else: attn_processor_class = ( IPAdapterAttnProcessor2_0 if hasattr(F, "scaled_dot_product_attention") else IPAdapterAttnProcessor ) attn_procs[name] = attn_processor_class( hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, scale=1.0, num_tokens=num_image_text_embeds, ).to(dtype=self.dtype, device=self.device) lora_dict.update( {f"unet.{name}.to_k_lora.down.weight": state_dict["ip_adapter"][f"{key_id}.to_k_lora.down.weight"]} ) lora_dict.update( {f"unet.{name}.to_q_lora.down.weight": state_dict["ip_adapter"][f"{key_id}.to_q_lora.down.weight"]} ) lora_dict.update( {f"unet.{name}.to_v_lora.down.weight": state_dict["ip_adapter"][f"{key_id}.to_v_lora.down.weight"]} ) lora_dict.update( { f"unet.{name}.to_out_lora.down.weight": state_dict["ip_adapter"][ f"{key_id}.to_out_lora.down.weight" ] } ) lora_dict.update( {f"unet.{name}.to_k_lora.up.weight": state_dict["ip_adapter"][f"{key_id}.to_k_lora.up.weight"]} ) lora_dict.update( {f"unet.{name}.to_q_lora.up.weight": state_dict["ip_adapter"][f"{key_id}.to_q_lora.up.weight"]} ) lora_dict.update( {f"unet.{name}.to_v_lora.up.weight": state_dict["ip_adapter"][f"{key_id}.to_v_lora.up.weight"]} ) lora_dict.update( {f"unet.{name}.to_out_lora.up.weight": state_dict["ip_adapter"][f"{key_id}.to_out_lora.up.weight"]} ) value_dict = {} value_dict.update({"to_k_ip.0.weight": state_dict["ip_adapter"][f"{key_id}.to_k_ip.weight"]}) value_dict.update({"to_v_ip.0.weight": state_dict["ip_adapter"][f"{key_id}.to_v_ip.weight"]}) attn_procs[name].load_state_dict(value_dict) key_id += 1 self.unet.set_attn_processor(attn_procs) self.load_lora_weights(lora_dict, adapter_name="faceid") self.set_adapters(["faceid"], adapter_weights=[1.0]) # convert IP-Adapter Image Projection layers to diffusers image_projection = self.convert_ip_adapter_image_proj_to_diffusers(state_dict["image_proj"]) image_projection_layers = [image_projection.to(device=self.device, dtype=self.dtype)] self.unet.encoder_hid_proj = MultiIPAdapterImageProjection(image_projection_layers) self.unet.config.encoder_hid_dim_type = "ip_image_proj" def set_ip_adapter_scale(self, scale): unet = getattr(self, self.unet_name) if not hasattr(self, "unet") else self.unet for attn_processor in unet.attn_processors.values(): if isinstance(attn_processor, (IPAdapterAttnProcessor, IPAdapterAttnProcessor2_0)): attn_processor.scale = [scale] def _encode_prompt( self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt=None, prompt_embeds: Optional[torch.Tensor] = None, negative_prompt_embeds: Optional[torch.Tensor] = None, lora_scale: Optional[float] = None, **kwargs, ): deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple." deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False) prompt_embeds_tuple = self.encode_prompt( prompt=prompt, device=device, num_images_per_prompt=num_images_per_prompt, do_classifier_free_guidance=do_classifier_free_guidance, negative_prompt=negative_prompt, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, lora_scale=lora_scale, **kwargs, ) # concatenate for backwards comp prompt_embeds = torch.cat([prompt_embeds_tuple[1], prompt_embeds_tuple[0]]) return prompt_embeds def encode_prompt( self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt=None, prompt_embeds: Optional[torch.Tensor] = None, negative_prompt_embeds: Optional[torch.Tensor] = None, lora_scale: Optional[float] = None, clip_skip: Optional[int] = 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.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. lora_scale (`float`, *optional*): A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. clip_skip (`int`, *optional*): Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that the output of the pre-final layer will be used for computing the prompt embeddings. """ # 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, StableDiffusionLoraLoaderMixin): self._lora_scale = lora_scale # dynamically adjust the LoRA scale if not USE_PEFT_BACKEND: adjust_lora_scale_text_encoder(self.text_encoder, lora_scale) else: scale_lora_layers(self.text_encoder, 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] if prompt_embeds is None: # textual inversion: process multi-vector tokens if necessary if isinstance(self, TextualInversionLoaderMixin): prompt = self.maybe_convert_prompt(prompt, self.tokenizer) text_inputs = self.tokenizer( prompt, padding="max_length", max_length=self.tokenizer.model_max_length, truncation=True, return_tensors="pt", ) text_input_ids = text_inputs.input_ids untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal( text_input_ids, untruncated_ids ): removed_text = self.tokenizer.batch_decode( untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1] ) logger.warning( "The following part of your input was truncated because CLIP can only handle sequences up to" f" {self.tokenizer.model_max_length} tokens: {removed_text}" ) if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask: attention_mask = text_inputs.attention_mask.to(device) else: attention_mask = None if clip_skip is None: prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask) prompt_embeds = prompt_embeds[0] else: prompt_embeds = self.text_encoder( text_input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True ) # Access the `hidden_states` first, that contains a tuple of # all the hidden states from the encoder layers. Then index into # the tuple to access the hidden states from the desired layer. prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)] # We also need to apply the final LayerNorm here to not mess with the # representations. The `last_hidden_states` that we typically use for # obtaining the final prompt representations passes through the LayerNorm # layer. prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds) if self.text_encoder is not None: prompt_embeds_dtype = self.text_encoder.dtype elif self.unet is not None: prompt_embeds_dtype = self.unet.dtype else: prompt_embeds_dtype = prompt_embeds.dtype prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device) bs_embed, seq_len, _ = prompt_embeds.shape # duplicate text embeddings for each generation per prompt, using mps friendly method 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) # get unconditional embeddings for classifier free guidance if 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 # textual inversion: process multi-vector tokens if necessary if isinstance(self, TextualInversionLoaderMixin): uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer) max_length = prompt_embeds.shape[1] uncond_input = self.tokenizer( uncond_tokens, padding="max_length", max_length=max_length, truncation=True, return_tensors="pt", ) if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask: attention_mask = uncond_input.attention_mask.to(device) else: attention_mask = None negative_prompt_embeds = self.text_encoder( uncond_input.input_ids.to(device), attention_mask=attention_mask, ) negative_prompt_embeds = negative_prompt_embeds[0] 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=prompt_embeds_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) if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND: # Retrieve the original scale by scaling back the LoRA layers unscale_lora_layers(self.text_encoder, lora_scale) return prompt_embeds, negative_prompt_embeds def run_safety_checker(self, image, device, dtype): if self.safety_checker is None: has_nsfw_concept = None else: if torch.is_tensor(image): feature_extractor_input = self.image_processor.postprocess(image, output_type="pil") else: feature_extractor_input = self.image_processor.numpy_to_pil(image) safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device) image, has_nsfw_concept = self.safety_checker( images=image, clip_input=safety_checker_input.pixel_values.to(dtype) ) return image, has_nsfw_concept def decode_latents(self, latents): deprecation_message = "The decode_latents method is deprecated and will be removed in 1.0.0. Please use VaeImageProcessor.postprocess(...) instead" deprecate("decode_latents", "1.0.0", deprecation_message, standard_warn=False) latents = 1 / self.vae.config.scaling_factor * latents image = self.vae.decode(latents, return_dict=False)[0] image = (image / 2 + 0.5).clamp(0, 1) # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16 image = image.cpu().permute(0, 2, 3, 1).float().numpy() return image 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, height, width, callback_steps, negative_prompt=None, prompt_embeds=None, negative_prompt_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 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}." ) def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None): 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 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.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32): """ See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298 Args: timesteps (`torch.Tensor`): generate embedding vectors at these timesteps embedding_dim (`int`, *optional*, defaults to 512): dimension of the embeddings to generate dtype: data type of the generated embeddings Returns: `torch.Tensor`: Embedding vectors with shape `(len(timesteps), 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 @property def clip_skip(self): return self._clip_skip # 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 num_timesteps(self): return self._num_timesteps @property def interrupt(self): return self._interrupt @torch.no_grad() def __call__( self, prompt: Union[str, List[str]] = None, height: Optional[int] = None, width: Optional[int] = None, num_inference_steps: int = 50, timesteps: List[int] = 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, image_embeds: Optional[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, clip_skip: Optional[int] = None, callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None, callback_on_step_end_tensor_inputs: List[str] = ["latents"], **kwargs, ): r""" The call function to the pipeline for generation. Args: prompt (`str` or `List[str]`, *optional*): The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`. height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): The height in pixels of the generated image. width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): The width in pixels of the generated 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. 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. guidance_scale (`float`, *optional*, defaults to 7.5): A higher guidance scale value encourages the model to generate images closely linked to the text `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`. negative_prompt (`str` or `List[str]`, *optional*): The prompt or prompts to guide what to not include in image generation. If not defined, you need to pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 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) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers. generator (`torch.Generator` or `List[torch.Generator]`, *optional*): A [`torch.Generator`](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 is 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 (prompt weighting). If not provided, text embeddings are generated from the `prompt` input argument. negative_prompt_embeds (`torch.Tensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument. image_embeds (`torch.Tensor`, *optional*): Pre-generated image embeddings. output_type (`str`, *optional*, defaults to `"pil"`): The output format of the generated image. Choose between `PIL.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 in [`self.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 from [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. clip_skip (`int`, *optional*): Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that the output of the pre-final layer will be used for computing the prompt embeddings. callback_on_step_end (`Callable`, *optional*): A function that calls at the end of each denoising steps during the inference. The function is called 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`: If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned, otherwise a `tuple` is returned where the first element is a list with the generated images and the second element is a list of `bool`s indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content. """ 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`", ) # 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 # to deal with lora scaling and other possible forward hooks # 1. Check inputs. Raise error if not correct self.check_inputs( prompt, height, width, callback_steps, negative_prompt, prompt_embeds, negative_prompt_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._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 = 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, lora_scale=lora_scale, clip_skip=self.clip_skip, ) # 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 if self.do_classifier_free_guidance: prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds]) if image_embeds is not None: image_embeds = torch.stack([image_embeds] * num_images_per_prompt, dim=0).to( device=device, dtype=prompt_embeds.dtype ) negative_image_embeds = torch.zeros_like(image_embeds) if self.do_classifier_free_guidance: image_embeds = torch.cat([negative_image_embeds, image_embeds]) image_embeds = [image_embeds] # 4. Prepare timesteps timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, 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. 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) # 6.1 Add image embeds for IP-Adapter added_cond_kwargs = {"image_embeds": image_embeds} if image_embeds is not None else {} # 6.2 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. Denoising loop num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order 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 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 = 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) # 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 not output_type == "latent": image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False, generator=generator)[ 0 ] image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype) else: image = latents has_nsfw_concept = None if has_nsfw_concept is None: do_denormalize = [True] * image.shape[0] else: do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept] image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize) # Offload all models self.maybe_free_model_hooks() if not return_dict: return (image, has_nsfw_concept) return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)