diffusers/community-pipelines-mirror

1# Copyright 2025 The HuggingFace Team. All rights reserved.2#3# Licensed under the Apache License, Version 2.0 (the "License");4# you may not use this file except in compliance with the License.5# You may obtain a copy of the License at6#7#     http://www.apache.org/licenses/LICENSE-2.08#9# Unless required by applicable law or agreed to in writing, software10# distributed under the License is distributed on an "AS IS" BASIS,11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.12# See the License for the specific language governing permissions and13# limitations under the License.14 15import inspect16from typing import Any, Callable, Dict, List, Optional, Union17 18import intel_extension_for_pytorch as ipex19import torch20from packaging import version21from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer22 23from diffusers.configuration_utils import FrozenDict24from diffusers.loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin25from diffusers.models import AutoencoderKL, UNet2DConditionModel26from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin27from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput28from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker29from diffusers.schedulers import KarrasDiffusionSchedulers30from diffusers.utils import (31    deprecate,32    logging,33    replace_example_docstring,34)35from diffusers.utils.torch_utils import randn_tensor36 37 38logger = logging.get_logger(__name__)  # pylint: disable=invalid-name39 40EXAMPLE_DOC_STRING = """41    Examples:42        ```py43        >>> import torch44        >>> from diffusers import StableDiffusionPipeline45 46        >>> pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", custom_pipeline="stable_diffusion_ipex")47 48        >>> # For Float3249        >>> pipe.prepare_for_ipex(prompt, dtype=torch.float32, height=512, width=512) #value of image height/width should be consistent with the pipeline inference50        >>> # For BFloat1651        >>> pipe.prepare_for_ipex(prompt, dtype=torch.bfloat16, height=512, width=512) #value of image height/width should be consistent with the pipeline inference52 53        >>> prompt = "a photo of an astronaut riding a horse on mars"54        >>> # For Float3255        >>> image = pipe(prompt, num_inference_steps=num_inference_steps, height=512, width=512).images[0] #value of image height/width should be consistent with 'prepare_for_ipex()'56        >>> # For BFloat1657        >>> with torch.cpu.amp.autocast(enabled=True, dtype=torch.bfloat16):58        >>>     image = pipe(prompt, num_inference_steps=num_inference_steps, height=512, width=512).images[0] #value of image height/width should be consistent with 'prepare_for_ipex()'59        ```60"""61 62 63class StableDiffusionIPEXPipeline(64    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin65):66    r"""67    Pipeline for text-to-image generation using Stable Diffusion on IPEX.68 69    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the70    library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)71 72    Args:73        vae ([`AutoencoderKL`]):74            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.75        text_encoder ([`CLIPTextModel`]):76            Frozen text-encoder. Stable Diffusion uses the text portion of77            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically78            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.79        tokenizer (`CLIPTokenizer`):80            Tokenizer of class81            [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).82        unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.83        scheduler ([`SchedulerMixin`]):84            A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of85            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].86        safety_checker ([`StableDiffusionSafetyChecker`]):87            Classification module that estimates whether generated images could be considered offensive or harmful.88            Please, refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for details.89        feature_extractor ([`CLIPImageProcessor`]):90            Model that extracts features from generated images to be used as inputs for the `safety_checker`.91    """92 93    _optional_components = ["safety_checker", "feature_extractor"]94 95    def __init__(96        self,97        vae: AutoencoderKL,98        text_encoder: CLIPTextModel,99        tokenizer: CLIPTokenizer,100        unet: UNet2DConditionModel,101        scheduler: KarrasDiffusionSchedulers,102        safety_checker: StableDiffusionSafetyChecker,103        feature_extractor: CLIPImageProcessor,104        requires_safety_checker: bool = True,105    ):106        super().__init__()107 108        if scheduler is not None and getattr(scheduler.config, "steps_offset", 1) != 1:109            deprecation_message = (110                f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`"111                f" should be set to 1 instead of {scheduler.config.steps_offset}. Please make sure "112                "to update the config accordingly as leaving `steps_offset` might led to incorrect results"113                " in future versions. If you have downloaded this checkpoint from the Hugging Face Hub,"114                " it would be very nice if you could open a Pull request for the `scheduler/scheduler_config.json`"115                " file"116            )117            deprecate("steps_offset!=1", "1.0.0", deprecation_message, standard_warn=False)118            new_config = dict(scheduler.config)119            new_config["steps_offset"] = 1120            scheduler._internal_dict = FrozenDict(new_config)121 122        if scheduler is not None and getattr(scheduler.config, "clip_sample", False) is True:123            deprecation_message = (124                f"The configuration file of this scheduler: {scheduler} has not set the configuration `clip_sample`."125                " `clip_sample` should be set to False in the configuration file. Please make sure to update the"126                " config accordingly as not setting `clip_sample` in the config might lead to incorrect results in"127                " future versions. If you have downloaded this checkpoint from the Hugging Face Hub, it would be very"128                " nice if you could open a Pull request for the `scheduler/scheduler_config.json` file"129            )130            deprecate("clip_sample not set", "1.0.0", deprecation_message, standard_warn=False)131            new_config = dict(scheduler.config)132            new_config["clip_sample"] = False133            scheduler._internal_dict = FrozenDict(new_config)134 135        if safety_checker is None and requires_safety_checker:136            logger.warning(137                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"138                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"139                " results in services or applications open to the public. Both the diffusers team and Hugging Face"140                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"141                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"142                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."143            )144 145        if safety_checker is not None and feature_extractor is None:146            raise ValueError(147                "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"148                " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."149            )150 151        is_unet_version_less_0_9_0 = (152            unet is not None153            and hasattr(unet.config, "_diffusers_version")154            and version.parse(version.parse(unet.config._diffusers_version).base_version) < version.parse("0.9.0.dev0")155        )156        is_unet_sample_size_less_64 = (157            unet is not None and hasattr(unet.config, "sample_size") and unet.config.sample_size < 64158        )159        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:160            deprecation_message = (161                "The configuration file of the unet has set the default `sample_size` to smaller than"162                " 64 which seems highly unlikely. If your checkpoint is a fine-tuned version of any of the"163                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"164                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"165                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"166                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"167                " in the config might lead to incorrect results in future versions. If you have downloaded this"168                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"169                " the `unet/config.json` file"170            )171            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)172            new_config = dict(unet.config)173            new_config["sample_size"] = 64174            unet._internal_dict = FrozenDict(new_config)175 176        self.register_modules(177            vae=vae,178            text_encoder=text_encoder,179            tokenizer=tokenizer,180            unet=unet,181            scheduler=scheduler,182            safety_checker=safety_checker,183            feature_extractor=feature_extractor,184        )185        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8186        self.register_to_config(requires_safety_checker=requires_safety_checker)187 188    def get_input_example(self, prompt, height=None, width=None, guidance_scale=7.5, num_images_per_prompt=1):189        prompt_embeds = None190        negative_prompt_embeds = None191        negative_prompt = None192        callback_steps = 1193        generator = None194        latents = None195 196        # 0. Default height and width to unet197        height = height or self.unet.config.sample_size * self.vae_scale_factor198        width = width or self.unet.config.sample_size * self.vae_scale_factor199 200        # 1. Check inputs. Raise error if not correct201        self.check_inputs(202            prompt, height, width, callback_steps, negative_prompt, prompt_embeds, negative_prompt_embeds203        )204 205        # 2. Define call parameters206        if prompt is not None and isinstance(prompt, str):207            batch_size = 1208        elif prompt is not None and isinstance(prompt, list):209            batch_size = len(prompt)210 211        device = "cpu"212        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)213        # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`214        # corresponds to doing no classifier free guidance.215        do_classifier_free_guidance = guidance_scale > 1.0216 217        # 3. Encode input prompt218        prompt_embeds = self._encode_prompt(219            prompt,220            device,221            num_images_per_prompt,222            do_classifier_free_guidance,223            negative_prompt,224            prompt_embeds=prompt_embeds,225            negative_prompt_embeds=negative_prompt_embeds,226        )227 228        # 5. Prepare latent variables229        latents = self.prepare_latents(230            batch_size * num_images_per_prompt,231            self.unet.config.in_channels,232            height,233            width,234            prompt_embeds.dtype,235            device,236            generator,237            latents,238        )239        dummy = torch.ones(1, dtype=torch.int32)240        latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents241        latent_model_input = self.scheduler.scale_model_input(latent_model_input, dummy)242 243        unet_input_example = (latent_model_input, dummy, prompt_embeds)244        vae_decoder_input_example = latents245 246        return unet_input_example, vae_decoder_input_example247 248    def prepare_for_ipex(self, promt, dtype=torch.float32, height=None, width=None, guidance_scale=7.5):249        self.unet = self.unet.to(memory_format=torch.channels_last)250        self.vae.decoder = self.vae.decoder.to(memory_format=torch.channels_last)251        self.text_encoder = self.text_encoder.to(memory_format=torch.channels_last)252        if self.safety_checker is not None:253            self.safety_checker = self.safety_checker.to(memory_format=torch.channels_last)254 255        unet_input_example, vae_decoder_input_example = self.get_input_example(promt, height, width, guidance_scale)256 257        # optimize with ipex258        if dtype == torch.bfloat16:259            self.unet = ipex.optimize(self.unet.eval(), dtype=torch.bfloat16, inplace=True)260            self.vae.decoder = ipex.optimize(self.vae.decoder.eval(), dtype=torch.bfloat16, inplace=True)261            self.text_encoder = ipex.optimize(self.text_encoder.eval(), dtype=torch.bfloat16, inplace=True)262            if self.safety_checker is not None:263                self.safety_checker = ipex.optimize(self.safety_checker.eval(), dtype=torch.bfloat16, inplace=True)264        elif dtype == torch.float32:265            self.unet = ipex.optimize(266                self.unet.eval(),267                dtype=torch.float32,268                inplace=True,269                weights_prepack=True,270                auto_kernel_selection=False,271            )272            self.vae.decoder = ipex.optimize(273                self.vae.decoder.eval(),274                dtype=torch.float32,275                inplace=True,276                weights_prepack=True,277                auto_kernel_selection=False,278            )279            self.text_encoder = ipex.optimize(280                self.text_encoder.eval(),281                dtype=torch.float32,282                inplace=True,283                weights_prepack=True,284                auto_kernel_selection=False,285            )286            if self.safety_checker is not None:287                self.safety_checker = ipex.optimize(288                    self.safety_checker.eval(),289                    dtype=torch.float32,290                    inplace=True,291                    weights_prepack=True,292                    auto_kernel_selection=False,293                )294        else:295            raise ValueError(" The value of 'dtype' should be 'torch.bfloat16' or 'torch.float32' !")296 297        # trace unet model to get better performance on IPEX298        with torch.cpu.amp.autocast(enabled=dtype == torch.bfloat16), torch.no_grad():299            unet_trace_model = torch.jit.trace(self.unet, unet_input_example, check_trace=False, strict=False)300            unet_trace_model = torch.jit.freeze(unet_trace_model)301        self.unet.forward = unet_trace_model.forward302 303        # trace vae.decoder model to get better performance on IPEX304        with torch.cpu.amp.autocast(enabled=dtype == torch.bfloat16), torch.no_grad():305            ave_decoder_trace_model = torch.jit.trace(306                self.vae.decoder, vae_decoder_input_example, check_trace=False, strict=False307            )308            ave_decoder_trace_model = torch.jit.freeze(ave_decoder_trace_model)309        self.vae.decoder.forward = ave_decoder_trace_model.forward310 311    def _encode_prompt(312        self,313        prompt,314        device,315        num_images_per_prompt,316        do_classifier_free_guidance,317        negative_prompt=None,318        prompt_embeds: Optional[torch.Tensor] = None,319        negative_prompt_embeds: Optional[torch.Tensor] = None,320    ):321        r"""322        Encodes the prompt into text encoder hidden states.323 324        Args:325             prompt (`str` or `List[str]`, *optional*):326                prompt to be encoded327            device: (`torch.device`):328                torch device329            num_images_per_prompt (`int`):330                number of images that should be generated per prompt331            do_classifier_free_guidance (`bool`):332                whether to use classifier free guidance or not333            negative_prompt (`str` or `List[str]`, *optional*):334                The prompt or prompts not to guide the image generation. If not defined, one has to pass335                `negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.336                Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).337            prompt_embeds (`torch.Tensor`, *optional*):338                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not339                provided, text embeddings will be generated from `prompt` input argument.340            negative_prompt_embeds (`torch.Tensor`, *optional*):341                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt342                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input343                argument.344        """345        if prompt is not None and isinstance(prompt, str):346            batch_size = 1347        elif prompt is not None and isinstance(prompt, list):348            batch_size = len(prompt)349        else:350            batch_size = prompt_embeds.shape[0]351 352        if prompt_embeds is None:353            # textual inversion: process multi-vector tokens if necessary354            if isinstance(self, TextualInversionLoaderMixin):355                prompt = self.maybe_convert_prompt(prompt, self.tokenizer)356 357            text_inputs = self.tokenizer(358                prompt,359                padding="max_length",360                max_length=self.tokenizer.model_max_length,361                truncation=True,362                return_tensors="pt",363            )364            text_input_ids = text_inputs.input_ids365            untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids366 367            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(368                text_input_ids, untruncated_ids369            ):370                removed_text = self.tokenizer.batch_decode(371                    untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]372                )373                logger.warning(374                    "The following part of your input was truncated because CLIP can only handle sequences up to"375                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"376                )377 378            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:379                attention_mask = text_inputs.attention_mask.to(device)380            else:381                attention_mask = None382 383            prompt_embeds = self.text_encoder(384                text_input_ids.to(device),385                attention_mask=attention_mask,386            )387            prompt_embeds = prompt_embeds[0]388 389        prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)390 391        bs_embed, seq_len, _ = prompt_embeds.shape392        # duplicate text embeddings for each generation per prompt, using mps friendly method393        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)394        prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)395 396        # get unconditional embeddings for classifier free guidance397        if do_classifier_free_guidance and negative_prompt_embeds is None:398            uncond_tokens: List[str]399            if negative_prompt is None:400                uncond_tokens = [""] * batch_size401            elif type(prompt) is not type(negative_prompt):402                raise TypeError(403                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="404                    f" {type(prompt)}."405                )406            elif isinstance(negative_prompt, str):407                uncond_tokens = [negative_prompt]408            elif batch_size != len(negative_prompt):409                raise ValueError(410                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"411                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"412                    " the batch size of `prompt`."413                )414            else:415                uncond_tokens = negative_prompt416 417            # textual inversion: process multi-vector tokens if necessary418            if isinstance(self, TextualInversionLoaderMixin):419                uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer)420 421            max_length = prompt_embeds.shape[1]422            uncond_input = self.tokenizer(423                uncond_tokens,424                padding="max_length",425                max_length=max_length,426                truncation=True,427                return_tensors="pt",428            )429 430            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:431                attention_mask = uncond_input.attention_mask.to(device)432            else:433                attention_mask = None434 435            negative_prompt_embeds = self.text_encoder(436                uncond_input.input_ids.to(device),437                attention_mask=attention_mask,438            )439            negative_prompt_embeds = negative_prompt_embeds[0]440 441        if do_classifier_free_guidance:442            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method443            seq_len = negative_prompt_embeds.shape[1]444 445            negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)446 447            negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)448            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)449 450            # For classifier free guidance, we need to do two forward passes.451            # Here we concatenate the unconditional and text embeddings into a single batch452            # to avoid doing two forward passes453            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])454 455        return prompt_embeds456 457    def run_safety_checker(self, image, device, dtype):458        if self.safety_checker is not None:459            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)460            image, has_nsfw_concept = self.safety_checker(461                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)462            )463        else:464            has_nsfw_concept = None465        return image, has_nsfw_concept466 467    def decode_latents(self, latents):468        latents = 1 / self.vae.config.scaling_factor * latents469        image = self.vae.decode(latents).sample470        image = (image / 2 + 0.5).clamp(0, 1)471        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16472        image = image.cpu().permute(0, 2, 3, 1).float().numpy()473        return image474 475    def prepare_extra_step_kwargs(self, generator, eta):476        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature477        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.478        # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502479        # and should be between [0, 1]480 481        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())482        extra_step_kwargs = {}483        if accepts_eta:484            extra_step_kwargs["eta"] = eta485 486        # check if the scheduler accepts generator487        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())488        if accepts_generator:489            extra_step_kwargs["generator"] = generator490        return extra_step_kwargs491 492    def check_inputs(493        self,494        prompt,495        height,496        width,497        callback_steps,498        negative_prompt=None,499        prompt_embeds=None,500        negative_prompt_embeds=None,501    ):502        if height % 8 != 0 or width % 8 != 0:503            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")504 505        if (callback_steps is None) or (506            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)507        ):508            raise ValueError(509                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"510                f" {type(callback_steps)}."511            )512 513        if prompt is not None and prompt_embeds is not None:514            raise ValueError(515                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"516                " only forward one of the two."517            )518        elif prompt is None and prompt_embeds is None:519            raise ValueError(520                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."521            )522        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):523            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")524 525        if negative_prompt is not None and negative_prompt_embeds is not None:526            raise ValueError(527                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"528                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."529            )530 531        if prompt_embeds is not None and negative_prompt_embeds is not None:532            if prompt_embeds.shape != negative_prompt_embeds.shape:533                raise ValueError(534                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"535                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"536                    f" {negative_prompt_embeds.shape}."537                )538 539    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):540        shape = (541            batch_size,542            num_channels_latents,543            int(height) // self.vae_scale_factor,544            int(width) // self.vae_scale_factor,545        )546        if isinstance(generator, list) and len(generator) != batch_size:547            raise ValueError(548                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"549                f" size of {batch_size}. Make sure the batch size matches the length of the generators."550            )551 552        if latents is None:553            latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)554        else:555            latents = latents.to(device)556 557        # scale the initial noise by the standard deviation required by the scheduler558        latents = latents * self.scheduler.init_noise_sigma559        return latents560 561    @torch.no_grad()562    @replace_example_docstring(EXAMPLE_DOC_STRING)563    def __call__(564        self,565        prompt: Union[str, List[str]] = None,566        height: Optional[int] = None,567        width: Optional[int] = None,568        num_inference_steps: int = 50,569        guidance_scale: float = 7.5,570        negative_prompt: Optional[Union[str, List[str]]] = None,571        num_images_per_prompt: Optional[int] = 1,572        eta: float = 0.0,573        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,574        latents: Optional[torch.Tensor] = None,575        prompt_embeds: Optional[torch.Tensor] = None,576        negative_prompt_embeds: Optional[torch.Tensor] = None,577        output_type: Optional[str] = "pil",578        return_dict: bool = True,579        callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,580        callback_steps: int = 1,581        cross_attention_kwargs: Optional[Dict[str, Any]] = None,582    ):583        r"""584        Function invoked when calling the pipeline for generation.585 586        Args:587            prompt (`str` or `List[str]`, *optional*):588                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.589                instead.590            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):591                The height in pixels of the generated image.592            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):593                The width in pixels of the generated image.594            num_inference_steps (`int`, *optional*, defaults to 50):595                The number of denoising steps. More denoising steps usually lead to a higher quality image at the596                expense of slower inference.597            guidance_scale (`float`, *optional*, defaults to 7.5):598                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://huggingface.co/papers/2207.12598).599                `guidance_scale` is defined as `w` of equation 2. of [Imagen600                Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting `guidance_scale >601                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,602                usually at the expense of lower image quality.603            negative_prompt (`str` or `List[str]`, *optional*):604                The prompt or prompts not to guide the image generation. If not defined, one has to pass605                `negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.606                Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).607            num_images_per_prompt (`int`, *optional*, defaults to 1):608                The number of images to generate per prompt.609            eta (`float`, *optional*, defaults to 0.0):610                Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only applies to611                [`schedulers.DDIMScheduler`], will be ignored for others.612            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):613                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)614                to make generation deterministic.615            latents (`torch.Tensor`, *optional*):616                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image617                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents618                tensor will ge generated by sampling using the supplied random `generator`.619            prompt_embeds (`torch.Tensor`, *optional*):620                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not621                provided, text embeddings will be generated from `prompt` input argument.622            negative_prompt_embeds (`torch.Tensor`, *optional*):623                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt624                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input625                argument.626            output_type (`str`, *optional*, defaults to `"pil"`):627                The output format of the generate image. Choose between628                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.629            return_dict (`bool`, *optional*, defaults to `True`):630                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a631                plain tuple.632            callback (`Callable`, *optional*):633                A function that will be called every `callback_steps` steps during inference. The function will be634                called with the following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.635            callback_steps (`int`, *optional*, defaults to 1):636                The frequency at which the `callback` function will be called. If not specified, the callback will be637                called at every step.638            cross_attention_kwargs (`dict`, *optional*):639                A kwargs dictionary that if specified is passed along to the `AttnProcessor` as defined under640                `self.processor` in641                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).642 643        Examples:644 645        Returns:646            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:647            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.648            When returning a tuple, the first element is a list with the generated images, and the second element is a649            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"650            (nsfw) content, according to the `safety_checker`.651        """652        # 0. Default height and width to unet653        height = height or self.unet.config.sample_size * self.vae_scale_factor654        width = width or self.unet.config.sample_size * self.vae_scale_factor655 656        # 1. Check inputs. Raise error if not correct657        self.check_inputs(658            prompt, height, width, callback_steps, negative_prompt, prompt_embeds, negative_prompt_embeds659        )660 661        # 2. Define call parameters662        if prompt is not None and isinstance(prompt, str):663            batch_size = 1664        elif prompt is not None and isinstance(prompt, list):665            batch_size = len(prompt)666        else:667            batch_size = prompt_embeds.shape[0]668 669        device = self._execution_device670        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)671        # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`672        # corresponds to doing no classifier free guidance.673        do_classifier_free_guidance = guidance_scale > 1.0674 675        # 3. Encode input prompt676        prompt_embeds = self._encode_prompt(677            prompt,678            device,679            num_images_per_prompt,680            do_classifier_free_guidance,681            negative_prompt,682            prompt_embeds=prompt_embeds,683            negative_prompt_embeds=negative_prompt_embeds,684        )685 686        # 4. Prepare timesteps687        self.scheduler.set_timesteps(num_inference_steps, device=device)688        timesteps = self.scheduler.timesteps689 690        # 5. Prepare latent variables691        num_channels_latents = self.unet.config.in_channels692        latents = self.prepare_latents(693            batch_size * num_images_per_prompt,694            num_channels_latents,695            height,696            width,697            prompt_embeds.dtype,698            device,699            generator,700            latents,701        )702 703        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline704        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)705 706        # 7. Denoising loop707        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order708        with self.progress_bar(total=num_inference_steps) as progress_bar:709            for i, t in enumerate(timesteps):710                # expand the latents if we are doing classifier free guidance711                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents712                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)713 714                # predict the noise residual715                noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=prompt_embeds)["sample"]716 717                # perform guidance718                if do_classifier_free_guidance:719                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)720                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)721 722                # compute the previous noisy sample x_t -> x_t-1723                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample724 725                # call the callback, if provided726                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):727                    progress_bar.update()728                    if callback is not None and i % callback_steps == 0:729                        step_idx = i // getattr(self.scheduler, "order", 1)730                        callback(step_idx, t, latents)731 732        if output_type == "latent":733            image = latents734            has_nsfw_concept = None735        elif output_type == "pil":736            # 8. Post-processing737            image = self.decode_latents(latents)738 739            # 9. Run safety checker740            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)741 742            # 10. Convert to PIL743            image = self.numpy_to_pil(image)744        else:745            # 8. Post-processing746            image = self.decode_latents(latents)747 748            # 9. Run safety checker749            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)750 751        # Offload last model to CPU752        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:753            self.final_offload_hook.offload()754 755        if not return_dict:756            return (image, has_nsfw_concept)757 758        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)759