# Adapted from Vchitect # This source code is licensed under the license found in the # LICENSE file in the root directory of this source tree. # -------------------------------------------------------- # References: # Vchitect: https://github.com/Vchitect/Vchitect-2.0 # diffusers: https://github.com/huggingface/diffusers # -------------------------------------------------------- import inspect import math from typing import Any, Callable, Dict, List, Optional, Union import torch import torch.distributed as dist from diffusers.image_processor import VaeImageProcessor from diffusers.models.autoencoders import AutoencoderKL from diffusers.schedulers import FlowMatchEulerDiscreteScheduler from diffusers.utils.torch_utils import randn_tensor from torch.amp import autocast from tqdm import tqdm from transformers import CLIPTextModelWithProjection, CLIPTokenizer, T5EncoderModel, T5TokenizerFast from videosys.core.pab_mgr import PABConfig, set_pab_manager, update_steps from videosys.core.pipeline import VideoSysPipeline, VideoSysPipelineOutput from videosys.models.transformers.vchitect_transformer_3d import VchitectXLTransformerModel from videosys.utils.logging import logger from videosys.utils.utils import save_video, set_seed class VchitectPABConfig(PABConfig): def __init__( self, spatial_broadcast: bool = True, spatial_threshold: list = [100, 800], spatial_range: int = 2, temporal_broadcast: bool = True, temporal_threshold: list = [100, 800], temporal_range: int = 4, cross_broadcast: bool = True, cross_threshold: list = [100, 800], cross_range: int = 6, ): super().__init__( spatial_broadcast=spatial_broadcast, spatial_threshold=spatial_threshold, spatial_range=spatial_range, temporal_broadcast=temporal_broadcast, temporal_threshold=temporal_threshold, temporal_range=temporal_range, cross_broadcast=cross_broadcast, cross_threshold=cross_threshold, cross_range=cross_range, ) class VchitectConfig: """ This config is to instantiate a `VchitectXLPipeline` class for video generation. To be specific, this config will be passed to engine by `VideoSysEngine(config)`. In the engine, it will be used to instantiate the corresponding pipeline class. And the engine will call the `generate` function of the pipeline to generate the video. If you want to explore the detail of generation, please refer to the pipeline class below. Args: model_path (str): The model path to use. Defaults to "Vchitect/Vchitect-2.0-2B". num_gpus (int): The number of GPUs to use. Defaults to 1. cpu_offload (bool): Whether to enable cpu offload. Defaults to False. enable_pab (bool): Whether to enable Pyramid Attention Broadcast. Defaults to False. pab_config (VchitectPABConfig): The configuration for Pyramid Attention Broadcast. Defaults to `VchitectPABConfig()`. Examples: ```python from videosys import OpenSoraPlanConfig, VideoSysEngine # change num_gpus for multi-gpu inference config = VchitectConfig("Vchitect/Vchitect-2.0-2B", num_gpus=1) engine = VideoSysEngine(config) prompt = "Sunset over the sea." # seed=-1 means random seed. >0 means fixed seed. # WxH: 480x288 624x352 432x240 768x432 video = engine.generate( prompt=prompt, negative_prompt="", num_inference_steps=100, guidance_scale=7.5, width=480, height=288, frames=40, seed=0, ).video[0] engine.save_video(video, f"./outputs/{prompt}.mp4") ``` """ def __init__( self, model_path: str = "Vchitect/Vchitect-2.0-2B", # ======= distributed ======== num_gpus: int = 1, # ======= memory ======= cpu_offload: bool = False, # ======= pab ======== enable_pab: bool = False, pab_config: VchitectPABConfig = VchitectPABConfig(), ): self.model_path = model_path self.pipeline_cls = VchitectXLPipeline # ======= distributed ======== self.num_gpus = num_gpus # ======= memory ======== self.cpu_offload = cpu_offload # ======= pab ======== self.enable_pab = enable_pab self.pab_config = pab_config class VchitectXLPipeline(VideoSysPipeline): r""" Args: transformer ([`VchitectXLTransformerModel`]): Conditional Transformer (MMDiT) architecture to denoise the encoded image latents. scheduler ([`FlowMatchEulerDiscreteScheduler`]): A scheduler to be used in combination with `transformer` to denoise the encoded image latents. vae ([`AutoencoderKL`]): Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. text_encoder ([`CLIPTextModelWithProjection`]): [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection), specifically the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant, with an additional added projection layer that is initialized with a diagonal matrix with the `hidden_size` as its dimension. text_encoder_2 ([`CLIPTextModelWithProjection`]): [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection), specifically the [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k) variant. text_encoder_3 ([`T5EncoderModel`]): Frozen text-encoder. Stable Diffusion 3 uses [T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel), specifically the [t5-v1_1-xxl](https://huggingface.co/google/t5-v1_1-xxl) variant. tokenizer (`CLIPTokenizer`): Tokenizer of class [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer). tokenizer_2 (`CLIPTokenizer`): Second Tokenizer of class [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer). tokenizer_3 (`T5TokenizerFast`): Tokenizer of class [T5Tokenizer](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Tokenizer). """ model_cpu_offload_seq = "text_encoder->text_encoder_2->text_encoder_3->transformer->vae" _optional_components = [ "text_encoder", "text_encoder_2", "text_encoder_3", "tokenizer", "tokenizer_2", "tokenizer_3", "vae", "transformer", "scheduler", ] _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds", "negative_pooled_prompt_embeds"] def __init__( self, config: VchitectConfig, text_encoder: Optional[CLIPTextModelWithProjection] = None, text_encoder_2: Optional[CLIPTextModelWithProjection] = None, text_encoder_3: Optional[T5EncoderModel] = None, tokenizer: Optional[CLIPTokenizer] = None, tokenizer_2: Optional[CLIPTokenizer] = None, tokenizer_3: Optional[T5TokenizerFast] = None, vae: Optional[AutoencoderKL] = None, transformer: Optional[VchitectXLTransformerModel] = None, scheduler: Optional[FlowMatchEulerDiscreteScheduler] = None, device: torch.device = torch.device("cuda"), dtype: torch.dtype = torch.bfloat16, ): super().__init__() self._config = config self._device = device self._dtype = dtype if text_encoder is None: self.text_encoder = CLIPTextModelWithProjection.from_pretrained( config.model_path, subfolder="text_encoder", torch_dtype=dtype ) if text_encoder_2 is None: self.text_encoder_2 = CLIPTextModelWithProjection.from_pretrained( config.model_path, subfolder="text_encoder_2", torch_dtype=dtype ) if text_encoder_3 is None: self.text_encoder_3 = T5EncoderModel.from_pretrained( config.model_path, subfolder="text_encoder_3", torch_dtype=dtype ) if tokenizer is None: self.tokenizer = CLIPTokenizer.from_pretrained(config.model_path, subfolder="tokenizer") if tokenizer_2 is None: self.tokenizer_2 = CLIPTokenizer.from_pretrained(config.model_path, subfolder="tokenizer_2") if tokenizer_3 is None: self.tokenizer_3 = T5TokenizerFast.from_pretrained(config.model_path, subfolder="tokenizer_3") if vae is None: self.vae = AutoencoderKL.from_pretrained(config.model_path, subfolder="vae", torch_dtype=dtype) if transformer is None: self.transformer = VchitectXLTransformerModel.from_pretrained( config.model_path, subfolder="transformer", torch_dtype=dtype ) if scheduler is None: self.scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained(config.model_path, subfolder="scheduler") self.register_modules( tokenizer=self.tokenizer, tokenizer_2=self.tokenizer_2, tokenizer_3=self.tokenizer_3, text_encoder=self.text_encoder, text_encoder_3=self.text_encoder_3, text_encoder_2=self.text_encoder_2, vae=self.vae, transformer=self.transformer, scheduler=self.scheduler, ) # pab if config.enable_pab: set_pab_manager(config.pab_config) # cpu offload if config.cpu_offload: self.enable_model_cpu_offload() else: self.set_eval_and_device( device, self.text_encoder, self.text_encoder_2, self.text_encoder_3, self.transformer, self.vae ) self.vae_scale_factor = ( 2 ** (len(self.vae.config.block_out_channels) - 1) if hasattr(self, "vae") and self.vae is not None else 8 ) self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor) self.tokenizer_max_length = ( self.tokenizer.model_max_length if hasattr(self, "tokenizer") and self.tokenizer is not None else 77 ) self.max_sequence_length_t5 = 256 self.default_sample_size = ( self.transformer.config.sample_size if hasattr(self, "transformer") and self.transformer is not None else 128 ) # parallel self._set_parallel() def _set_parallel( self, dp_size: Optional[int] = None, sp_size: Optional[int] = None, enable_cp: Optional[bool] = False ): # init sequence parallel if sp_size is None: sp_size = dist.get_world_size() dp_size = 1 else: assert ( dist.get_world_size() % sp_size == 0 ), f"world_size {dist.get_world_size()} must be divisible by sp_size" dp_size = dist.get_world_size() // sp_size # transformer parallel self.transformer.enable_parallel(dp_size, sp_size, enable_cp) def _get_t5_prompt_embeds( self, prompt: Union[str, List[str]] = None, num_images_per_prompt: int = 1, device: Optional[torch.device] = None, dtype: Optional[torch.dtype] = None, ): device = device or self.execution_device dtype = dtype or self.text_encoder.dtype prompt = [prompt] if isinstance(prompt, str) else prompt batch_size = len(prompt) if self.text_encoder_3 is None: return torch.zeros( (batch_size, self.max_sequence_length_t5, self.transformer.config.joint_attention_dim), device=device, dtype=dtype, ) text_inputs = self.tokenizer_3( prompt, padding="max_length", max_length=self.max_sequence_length_t5, truncation=True, add_special_tokens=True, return_tensors="pt", ) text_input_ids = text_inputs.input_ids untruncated_ids = self.tokenizer_3(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_3.batch_decode(untruncated_ids[:, self.max_sequence_length_t5 - 1 : -1]) logger.warning( "The following part of your input was truncated because CLIP can only handle sequences up to" f" {self.max_sequence_length_t5} tokens: {removed_text}" ) prompt_embeds = self.text_encoder_3(text_input_ids.to(device))[0] dtype = self.text_encoder_3.dtype prompt_embeds = prompt_embeds.to(dtype=dtype, device=device) _, seq_len, _ = prompt_embeds.shape # duplicate text embeddings and attention mask 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(batch_size * num_images_per_prompt, seq_len, -1) return prompt_embeds def _get_clip_prompt_embeds( self, prompt: Union[str, List[str]], num_images_per_prompt: int = 1, device: Optional[torch.device] = None, clip_skip: Optional[int] = None, clip_model_index: int = 0, ): device = device or self.execution_device clip_tokenizers = [self.tokenizer, self.tokenizer_2] clip_text_encoders = [self.text_encoder, self.text_encoder_2] tokenizer = clip_tokenizers[clip_model_index] text_encoder = clip_text_encoders[clip_model_index] prompt = [prompt] if isinstance(prompt, str) else prompt batch_size = len(prompt) text_inputs = tokenizer( prompt, padding="max_length", max_length=self.tokenizer_max_length, truncation=True, return_tensors="pt", ) text_input_ids = text_inputs.input_ids untruncated_ids = 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 = tokenizer.batch_decode(untruncated_ids[:, self.tokenizer_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_max_length} tokens: {removed_text}" ) prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True) pooled_prompt_embeds = prompt_embeds[0] if clip_skip is None: prompt_embeds = prompt_embeds.hidden_states[-2] else: prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)] prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device) _, 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(batch_size * num_images_per_prompt, seq_len, -1) pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt, 1) pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1) return prompt_embeds, pooled_prompt_embeds def _set_seed(self, seed): if dist.get_world_size() == 1: set_seed(seed) else: set_seed(seed, self.transformer.parallel_manager.dp_rank) @autocast("cuda", enabled=False) def encode_prompt( self, prompt: Union[str, List[str]], prompt_2: Union[str, List[str]], prompt_3: Union[str, List[str]], device: Optional[torch.device] = None, num_images_per_prompt: int = 1, do_classifier_free_guidance: bool = True, negative_prompt: Optional[Union[str, List[str]]] = None, negative_prompt_2: Optional[Union[str, List[str]]] = None, negative_prompt_3: Optional[Union[str, List[str]]] = None, prompt_embeds: Optional[torch.FloatTensor] = None, negative_prompt_embeds: Optional[torch.FloatTensor] = None, pooled_prompt_embeds: Optional[torch.FloatTensor] = None, negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, clip_skip: Optional[int] = None, ): r""" Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded 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 all text-encoders prompt_3 (`str` or `List[str]`, *optional*): The prompt or prompts to be sent to the `tokenizer_3` and `text_encoder_3`. If not defined, `prompt` is used in all text-encoders 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`). 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 all the text-encoders. negative_prompt_2 (`str` or `List[str]`, *optional*): The prompt or prompts not to guide the image generation to be sent to `tokenizer_3` and `text_encoder_3`. If not defined, `negative_prompt` is used in both text-encoders prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, text embeddings will be generated from `prompt` input argument. negative_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input argument. pooled_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, pooled text embeddings will be generated from `prompt` input argument. negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` input argument. 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. """ device = device or self.execution_device prompt = [prompt] if isinstance(prompt, str) else prompt if prompt is not None: batch_size = len(prompt) else: batch_size = prompt_embeds.shape[0] if prompt_embeds is None: prompt_2 = prompt_2 or prompt prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2 prompt_3 = prompt_3 or prompt prompt_3 = [prompt_3] if isinstance(prompt_3, str) else prompt_3 prompt_embed, pooled_prompt_embed = self._get_clip_prompt_embeds( prompt=prompt, device=device, num_images_per_prompt=num_images_per_prompt, clip_skip=clip_skip, clip_model_index=0, ) prompt_2_embed, pooled_prompt_2_embed = self._get_clip_prompt_embeds( prompt=prompt_2, device=device, num_images_per_prompt=num_images_per_prompt, clip_skip=clip_skip, clip_model_index=1, ) clip_prompt_embeds = torch.cat([prompt_embed, prompt_2_embed], dim=-1) t5_prompt_embed = self._get_t5_prompt_embeds( prompt=prompt_3, num_images_per_prompt=num_images_per_prompt, device=device, ) clip_prompt_embeds = torch.nn.functional.pad( clip_prompt_embeds, (0, t5_prompt_embed.shape[-1] - clip_prompt_embeds.shape[-1]) ) prompt_embeds = torch.cat([clip_prompt_embeds, t5_prompt_embed], dim=-2) pooled_prompt_embeds = torch.cat([pooled_prompt_embed, pooled_prompt_2_embed], dim=-1) if do_classifier_free_guidance and negative_prompt_embeds is None: negative_prompt = negative_prompt or "" negative_prompt_2 = negative_prompt_2 or negative_prompt negative_prompt_3 = negative_prompt_3 or negative_prompt # normalize str to list negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt negative_prompt_2 = ( batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2 ) negative_prompt_3 = ( batch_size * [negative_prompt_3] if isinstance(negative_prompt_3, str) else negative_prompt_3 ) if 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 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`." ) negative_prompt_embed, negative_pooled_prompt_embed = self._get_clip_prompt_embeds( negative_prompt, device=device, num_images_per_prompt=num_images_per_prompt, clip_skip=None, clip_model_index=0, ) negative_prompt_2_embed, negative_pooled_prompt_2_embed = self._get_clip_prompt_embeds( negative_prompt_2, device=device, num_images_per_prompt=num_images_per_prompt, clip_skip=None, clip_model_index=1, ) negative_clip_prompt_embeds = torch.cat([negative_prompt_embed, negative_prompt_2_embed], dim=-1) t5_negative_prompt_embed = self._get_t5_prompt_embeds( prompt=negative_prompt_3, num_images_per_prompt=num_images_per_prompt, device=device ) negative_clip_prompt_embeds = torch.nn.functional.pad( negative_clip_prompt_embeds, (0, t5_negative_prompt_embed.shape[-1] - negative_clip_prompt_embeds.shape[-1]), ) negative_prompt_embeds = torch.cat([negative_clip_prompt_embeds, t5_negative_prompt_embed], dim=-2) negative_pooled_prompt_embeds = torch.cat( [negative_pooled_prompt_embed, negative_pooled_prompt_2_embed], dim=-1 ) prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds = ( prompt_embeds.to(self._device), negative_prompt_embeds.to(self._device), pooled_prompt_embeds.to(self._device), negative_pooled_prompt_embeds.to(self._device), ) return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds def check_inputs( self, prompt, prompt_2, prompt_3, height, width, negative_prompt=None, negative_prompt_2=None, negative_prompt_3=None, prompt_embeds=None, negative_prompt_embeds=None, pooled_prompt_embeds=None, negative_pooled_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_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_3 is not None and prompt_embeds is not None: raise ValueError( f"Cannot forward both `prompt_3`: {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)}") elif prompt_3 is not None and (not isinstance(prompt_3, str) and not isinstance(prompt_3, list)): raise ValueError(f"`prompt_3` has to be of type `str` or `list` but is {type(prompt_3)}") 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." ) elif negative_prompt_3 is not None and negative_prompt_embeds is not None: raise ValueError( f"Cannot forward both `negative_prompt_3`: {negative_prompt_3} 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`." ) def prepare_latents( self, batch_size, num_channels_latents, height, width, frames, dtype, device, generator, latents=None, ): if latents is not None: return latents.to(device=device, dtype=dtype) # 1, 60, 16, 32, 32 shape = ( batch_size, frames, 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." ) latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype) return latents @property def guidance_scale(self): return self._guidance_scale @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 @property def joint_attention_kwargs(self): return self._joint_attention_kwargs @property def num_timesteps(self): return self._num_timesteps @property def interrupt(self): return self._interrupt @torch.no_grad() @autocast("cuda", dtype=torch.bfloat16) def generate( self, prompt: Union[str, List[str]] = None, prompt_2: Optional[Union[str, List[str]]] = None, prompt_3: Optional[Union[str, List[str]]] = None, height: int = 288, width: int = 480, frames: int = 40, num_inference_steps: int = 100, timesteps: List[int] = None, guidance_scale: float = 7.5, seed: int = -1, negative_prompt: Optional[Union[str, List[str]]] = None, negative_prompt_2: Optional[Union[str, List[str]]] = None, negative_prompt_3: Optional[Union[str, List[str]]] = None, num_images_per_prompt: Optional[int] = 1, generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, latents: Optional[torch.FloatTensor] = None, prompt_embeds: Optional[torch.FloatTensor] = None, negative_prompt_embeds: Optional[torch.FloatTensor] = None, pooled_prompt_embeds: Optional[torch.FloatTensor] = None, negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, output_type: Optional[str] = "pil", return_dict: bool = True, joint_attention_kwargs: Optional[Dict[str, Any]] = None, 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"], ): 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 `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is will be used instead prompt_3 (`str` or `List[str]`, *optional*): The prompt or prompts to be sent to `tokenizer_3` and `text_encoder_3`. If not defined, `prompt` is will be used instead 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. 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. 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 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 instead negative_prompt_3 (`str` or `List[str]`, *optional*): The prompt or prompts not to guide the image generation to be sent to `tokenizer_3` and `text_encoder_3`. If not defined, `negative_prompt` is used instead num_images_per_prompt (`int`, *optional*, defaults to 1): The number of images to generate per prompt. 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.FloatTensor`, *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.FloatTensor`, *optional*): Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, text embeddings will be generated from `prompt` input argument. negative_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input argument. pooled_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, pooled text embeddings will be generated from `prompt` input argument. negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` input argument. 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. joint_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). 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_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. """ height = height or self.default_sample_size * self.vae_scale_factor width = width or self.default_sample_size * self.vae_scale_factor frames = frames or 24 self._set_seed(seed) update_steps(num_inference_steps) # 1. Check inputs. Raise error if not correct self.check_inputs( prompt, prompt_2, prompt_3, height, width, negative_prompt=negative_prompt, negative_prompt_2=negative_prompt_2, negative_prompt_3=negative_prompt_3, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, pooled_prompt_embeds=pooled_prompt_embeds, negative_pooled_prompt_embeds=negative_pooled_prompt_embeds, callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs, ) self._guidance_scale = guidance_scale self._clip_skip = clip_skip self._joint_attention_kwargs = joint_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] ( prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds, ) = self.encode_prompt( prompt=prompt, prompt_2=prompt_2, prompt_3=prompt_3, negative_prompt=negative_prompt, negative_prompt_2=negative_prompt_2, negative_prompt_3=negative_prompt_3, do_classifier_free_guidance=self.do_classifier_free_guidance, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, pooled_prompt_embeds=pooled_prompt_embeds, negative_pooled_prompt_embeds=negative_pooled_prompt_embeds, device=self.text_encoder.device, clip_skip=self.clip_skip, num_images_per_prompt=num_images_per_prompt, ) if self.do_classifier_free_guidance: prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) pooled_prompt_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], dim=0) # 4. Prepare timesteps timesteps, num_inference_steps = retrieve_timesteps( self.scheduler, num_inference_steps, self._device, timesteps ) num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0) self._num_timesteps = len(timesteps) # 5. Prepare latent variables num_channels_latents = self.transformer.config.in_channels latents = self.prepare_latents( batch_size * num_images_per_prompt, num_channels_latents, height, width, frames, prompt_embeds.dtype, self._device, generator, latents, ) # 6. Denoising loop # with self.progress_bar(total=num_inference_steps) as progress_bar: for i, t in tqdm(enumerate(timesteps), total=len(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 # broadcast to batch dimension in a way that's compatible with ONNX/Core ML timestep = t.expand(latents.shape[0]) noise_pred_uncond = self.transformer( hidden_states=latent_model_input[0, :].unsqueeze(0), timestep=timestep, encoder_hidden_states=prompt_embeds[0, :].unsqueeze(0), pooled_projections=pooled_prompt_embeds[0, :].unsqueeze(0), joint_attention_kwargs=self.joint_attention_kwargs, return_dict=False, )[0] noise_pred_text = self.transformer( hidden_states=latent_model_input[1, :].unsqueeze(0), timestep=timestep, encoder_hidden_states=prompt_embeds[1, :].unsqueeze(0), pooled_projections=pooled_prompt_embeds[1, :].unsqueeze(0), joint_attention_kwargs=self.joint_attention_kwargs, return_dict=False, )[0] self._guidance_scale = 1 + guidance_scale * ( (1 - math.cos(math.pi * ((num_inference_steps - t.item()) / num_inference_steps) ** 5.0)) / 2 ) # 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) # compute the previous noisy sample x_t -> x_t-1 latents_dtype = latents.dtype latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0] if latents.dtype != latents_dtype: if torch.backends.mps.is_available(): # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272 latents = latents.to(latents_dtype) if 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) negative_pooled_prompt_embeds = callback_outputs.pop( "negative_pooled_prompt_embeds", negative_pooled_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 output_type == "latent": # image = latents # else: latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor videos = [] for v_idx in range(latents.shape[1]): image = self.vae.decode(latents[:, v_idx], return_dict=False)[0] image = self.image_processor.postprocess(image, output_type=output_type) videos.append(image[0]) videos = [videos] # Offload all models self.maybe_free_model_hooks() if not return_dict: return (videos,) return VideoSysPipelineOutput(video=videos) def save_video(self, video, output_path): save_video(video, output_path, fps=8) # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps def retrieve_timesteps( scheduler, num_inference_steps: Optional[int] = None, device: Optional[Union[str, torch.device]] = None, timesteps: Optional[List[int]] = None, sigmas: Optional[List[float]] = None, **kwargs, ): """ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. Args: scheduler (`SchedulerMixin`): The scheduler to get timesteps from. num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps` must be `None`. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. timesteps (`List[int]`, *optional*): Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed, `num_inference_steps` and `sigmas` must be `None`. sigmas (`List[float]`, *optional*): Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed, `num_inference_steps` and `timesteps` must be `None`. Returns: `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the second element is the number of inference steps. """ if timesteps is not None and sigmas is not None: raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values") if timesteps is not None: accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) if not accepts_timesteps: raise ValueError( f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" f" timestep schedules. Please check whether you are using the correct scheduler." ) scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs) timesteps = scheduler.timesteps num_inference_steps = len(timesteps) elif sigmas is not None: accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) if not accept_sigmas: raise ValueError( f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" f" sigmas schedules. Please check whether you are using the correct scheduler." ) scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs) timesteps = scheduler.timesteps num_inference_steps = len(timesteps) else: scheduler.set_timesteps(num_inference_steps, device=device, **kwargs) timesteps = scheduler.timesteps return timesteps, num_inference_steps