import inspect import math from typing import Any, Callable, Dict, List, Optional, Tuple, Union import torch import numpy as np import PIL from PIL import Image from torchvision import transforms from einops import rearrange, repeat from transformers import T5EncoderModel, T5Tokenizer from diffusers.video_processor import VideoProcessor from diffusers.utils.torch_utils import randn_tensor from diffusers.models.embeddings import get_3d_rotary_pos_embed from diffusers.pipelines.pipeline_utils import DiffusionPipeline from diffusers.models import AutoencoderKLCogVideoX, CogVideoXTransformer3DModel from diffusers import CogVideoXDDIMScheduler, CogVideoXDPMScheduler from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback from diffusers.pipelines.cogvideo.pipeline_cogvideox import CogVideoXPipelineOutput, CogVideoXLoraLoaderMixin from diffusers.image_processor import PipelineImageInput from cogvideo_controlnet_pcd import CogVideoXControlnetPCD as CogVideoXControlnet def resize_for_crop(image, crop_h, crop_w): img_h, img_w = image.shape[-2:] if img_h >= crop_h and img_w >= crop_w: coef = max(crop_h / img_h, crop_w / img_w) elif img_h <= crop_h and img_w <= crop_w: coef = max(crop_h / img_h, crop_w / img_w) else: coef = crop_h / img_h if crop_h > img_h else crop_w / img_w out_h, out_w = int(img_h * coef), int(img_w * coef) resized_image = transforms.functional.resize(image, (out_h, out_w), antialias=True) return resized_image def prepare_frames(input_images, video_size, do_resize=True, do_crop=True): input_images = np.stack([np.array(x) for x in input_images]) images_tensor = torch.from_numpy(input_images).permute(0, 3, 1, 2) / 127.5 - 1 if do_resize: images_tensor = [resize_for_crop(x, crop_h=video_size[0], crop_w=video_size[1]) for x in images_tensor] if do_crop: images_tensor = [transforms.functional.center_crop(x, video_size) for x in images_tensor] if isinstance(images_tensor, list): images_tensor = torch.stack(images_tensor) return images_tensor.unsqueeze(0) def get_resize_crop_region_for_grid(src, tgt_width, tgt_height): tw = tgt_width th = tgt_height h, w = src r = h / w if r > (th / tw): resize_height = th resize_width = int(round(th / h * w)) else: resize_width = tw resize_height = int(round(tw / w * h)) crop_top = int(round((th - resize_height) / 2.0)) crop_left = int(round((tw - resize_width) / 2.0)) return (crop_top, crop_left), (crop_top + resize_height, crop_left + resize_width) # 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 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents def retrieve_latents( encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample" ): if hasattr(encoder_output, "latent_dist") and sample_mode == "sample": return encoder_output.latent_dist.sample(generator) elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax": return encoder_output.latent_dist.mode() elif hasattr(encoder_output, "latents"): return encoder_output.latents else: raise AttributeError("Could not access latents of provided encoder_output") class ControlnetCogVideoXImageToVideoPCDPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin): _optional_components = [] model_cpu_offload_seq = "text_encoder->transformer->vae" _callback_tensor_inputs = [ "latents", "prompt_embeds", "negative_prompt_embeds", ] def __init__( self, tokenizer: T5Tokenizer, text_encoder: T5EncoderModel, vae: AutoencoderKLCogVideoX, transformer: CogVideoXTransformer3DModel, scheduler: Union[CogVideoXDDIMScheduler, CogVideoXDPMScheduler], controlnet: CogVideoXControlnet, ): super().__init__() self.register_modules( tokenizer=tokenizer, text_encoder=text_encoder, vae=vae, transformer=transformer, controlnet=controlnet, scheduler=scheduler ) self.vae_scale_factor_spatial = ( 2 ** (len(self.vae.config.block_out_channels) - 1) if hasattr(self, "vae") and self.vae is not None else 8 ) self.vae_scale_factor_temporal = ( self.vae.config.temporal_compression_ratio if hasattr(self, "vae") and self.vae is not None else 4 ) self.vae_scaling_factor_image = ( self.vae.config.scaling_factor if hasattr(self, "vae") and self.vae is not None else 0.7 ) self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial) # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline._get_t5_prompt_embeds def _get_t5_prompt_embeds( self, prompt: Union[str, List[str]] = None, num_videos_per_prompt: int = 1, max_sequence_length: int = 226, 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) text_inputs = self.tokenizer( prompt, padding="max_length", max_length=max_sequence_length, truncation=True, add_special_tokens=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[:, max_sequence_length - 1 : -1]) logger.warning( "The following part of your input was truncated because `max_sequence_length` is set to " f" {max_sequence_length} tokens: {removed_text}" ) prompt_embeds = self.text_encoder(text_input_ids.to(device))[0] prompt_embeds = prompt_embeds.to(dtype=dtype, device=device) # duplicate text embeddings for each generation per prompt, using mps friendly method _, seq_len, _ = prompt_embeds.shape prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1) prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1) return prompt_embeds # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.encode_prompt def encode_prompt( self, prompt: Union[str, List[str]], negative_prompt: Optional[Union[str, List[str]]] = None, do_classifier_free_guidance: bool = True, num_videos_per_prompt: int = 1, prompt_embeds: Optional[torch.Tensor] = None, negative_prompt_embeds: Optional[torch.Tensor] = None, max_sequence_length: int = 226, device: Optional[torch.device] = None, dtype: Optional[torch.dtype] = None, ): r""" Encodes the prompt into text encoder hidden states. Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded 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`). do_classifier_free_guidance (`bool`, *optional*, defaults to `True`): Whether to use classifier free guidance or not. num_videos_per_prompt (`int`, *optional*, defaults to 1): Number of videos that should be generated per prompt. torch device to place the resulting embeddings on 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. device: (`torch.device`, *optional*): torch device dtype: (`torch.dtype`, *optional*): torch dtype """ 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_embeds = self._get_t5_prompt_embeds( prompt=prompt, num_videos_per_prompt=num_videos_per_prompt, max_sequence_length=max_sequence_length, device=device, dtype=dtype, ) if do_classifier_free_guidance and negative_prompt_embeds is None: negative_prompt = negative_prompt or "" negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt 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_embeds = self._get_t5_prompt_embeds( prompt=negative_prompt, num_videos_per_prompt=num_videos_per_prompt, max_sequence_length=max_sequence_length, device=device, dtype=dtype, ) return prompt_embeds, negative_prompt_embeds def prepare_latents( self, image: torch.Tensor, batch_size: int = 1, num_channels_latents: int = 16, num_frames: int = 13, height: int = 60, width: int = 90, dtype: Optional[torch.dtype] = None, device: Optional[torch.device] = None, generator: Optional[torch.Generator] = None, latents: Optional[torch.Tensor] = None, ): 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." ) num_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1 shape = ( batch_size, num_frames, num_channels_latents, height // self.vae_scale_factor_spatial, width // self.vae_scale_factor_spatial, ) # For CogVideoX1.5, the latent should add 1 for padding (Not use) if self.transformer.config.patch_size_t is not None: shape = shape[:1] + (shape[1] + shape[1] % self.transformer.config.patch_size_t,) + shape[2:] image = image.unsqueeze(2) # [B, C, F, H, W] if isinstance(generator, list): image_latents = [ retrieve_latents(self.vae.encode(image[i].unsqueeze(0)), generator[i]) for i in range(batch_size) ] else: image_latents = [retrieve_latents(self.vae.encode(img.unsqueeze(0)), generator) for img in image] image_latents = torch.cat(image_latents, dim=0).to(dtype).permute(0, 2, 1, 3, 4) # [B, F, C, H, W] if not self.vae.config.invert_scale_latents: image_latents = self.vae_scaling_factor_image * image_latents else: # This is awkward but required because the CogVideoX team forgot to multiply the # scaling factor during training :) image_latents = 1 / self.vae_scaling_factor_image * image_latents padding_shape = ( batch_size, num_frames - 1, num_channels_latents, height // self.vae_scale_factor_spatial, width // self.vae_scale_factor_spatial, ) latent_padding = torch.zeros(padding_shape, device=device, dtype=dtype) image_latents = torch.cat([image_latents, latent_padding], dim=1) # Select the first frame along the second dimension if self.transformer.config.patch_size_t is not None: first_frame = image_latents[:, : image_latents.size(1) % self.transformer.config.patch_size_t, ...] image_latents = torch.cat([first_frame, image_latents], dim=1) 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, image_latents # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.decode_latents def decode_latents(self, latents: torch.Tensor) -> torch.Tensor: latents = latents.permute(0, 2, 1, 3, 4) # [batch_size, num_channels, num_frames, height, width] latents = 1 / self.vae_scaling_factor_image * latents frames = self.vae.decode(latents).sample return frames # Copied from diffusers.pipelines.animatediff.pipeline_animatediff_video2video.AnimateDiffVideoToVideoPipeline.get_timesteps def get_timesteps(self, num_inference_steps, timesteps, strength, device): # get the original timestep using init_timestep init_timestep = min(int(num_inference_steps * strength), num_inference_steps) t_start = max(num_inference_steps - init_timestep, 0) timesteps = timesteps[t_start * self.scheduler.order :] return timesteps, num_inference_steps - t_start # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs def prepare_extra_step_kwargs(self, generator, eta): # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 # and should be between [0, 1] accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) extra_step_kwargs = {} if accepts_eta: extra_step_kwargs["eta"] = eta # check if the scheduler accepts generator accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) if accepts_generator: extra_step_kwargs["generator"] = generator return extra_step_kwargs def check_inputs( self, image, prompt, height, width, negative_prompt, callback_on_step_end_tensor_inputs, prompt_embeds=None, negative_prompt_embeds=None, ): if ( not isinstance(image, torch.Tensor) and not isinstance(image, PIL.Image.Image) and not isinstance(image, list) ): raise ValueError( "`image` has to be of type `torch.Tensor` or `PIL.Image.Image` or `List[PIL.Image.Image]` but is" f" {type(image)}" ) 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 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 prompt is not None and negative_prompt_embeds is not None: raise ValueError( f"Cannot forward both `prompt`: {prompt} and `negative_prompt_embeds`:" f" {negative_prompt_embeds}. Please make sure to only forward one of the two." ) 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}." ) # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.fuse_qkv_projections def fuse_qkv_projections(self) -> None: r"""Enables fused QKV projections.""" self.fusing_transformer = True self.transformer.fuse_qkv_projections() # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.unfuse_qkv_projections def unfuse_qkv_projections(self) -> None: r"""Disable QKV projection fusion if enabled.""" if not self.fusing_transformer: logger.warning("The Transformer was not initially fused for QKV projections. Doing nothing.") else: self.transformer.unfuse_qkv_projections() self.fusing_transformer = False # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline._prepare_rotary_positional_embeddings def _prepare_rotary_positional_embeddings( self, height: int, width: int, num_frames: int, device: torch.device, ) -> Tuple[torch.Tensor, torch.Tensor]: grid_height = height // (self.vae_scale_factor_spatial * self.transformer.config.patch_size) grid_width = width // (self.vae_scale_factor_spatial * self.transformer.config.patch_size) p = self.transformer.config.patch_size p_t = self.transformer.config.patch_size_t base_size_width = self.transformer.config.sample_width // p base_size_height = self.transformer.config.sample_height // p if p_t is None: # CogVideoX 1.0 grid_crops_coords = get_resize_crop_region_for_grid( (grid_height, grid_width), base_size_width, base_size_height ) freqs_cos, freqs_sin = get_3d_rotary_pos_embed( embed_dim=self.transformer.config.attention_head_dim, crops_coords=grid_crops_coords, grid_size=(grid_height, grid_width), temporal_size=num_frames, device=device, ) else: # CogVideoX 1.5 base_num_frames = (num_frames + p_t - 1) // p_t freqs_cos, freqs_sin = get_3d_rotary_pos_embed( embed_dim=self.transformer.config.attention_head_dim, crops_coords=None, grid_size=(grid_height, grid_width), temporal_size=base_num_frames, grid_type="slice", max_size=(base_size_height, base_size_width), device=device, ) return freqs_cos, freqs_sin def encode_video(self, video): video = video.to(self.device, dtype=self.vae.dtype) video = video.permute(0, 2, 1, 3, 4) # [B, C, F, H, W] latent_dist = self.vae.encode(video).latent_dist.sample() * self.vae.config.scaling_factor return latent_dist.permute(0, 2, 1, 3, 4).to(memory_format=torch.contiguous_format) @property def guidance_scale(self): return self._guidance_scale @property def num_timesteps(self): return self._num_timesteps @property def attention_kwargs(self): return self._attention_kwargs @property def interrupt(self): return self._interrupt @torch.no_grad() def __call__( self, image: PipelineImageInput, anchor_video: None, controlnet_output_mask: None, prompt: Optional[Union[str, List[str]]] = None, negative_prompt: Optional[Union[str, List[str]]] = None, height: Optional[int] = None, width: Optional[int] = None, num_frames: int = 49, num_inference_steps: int = 50, timesteps: Optional[List[int]] = None, guidance_scale: float = 6, use_dynamic_cfg: bool = False, num_videos_per_prompt: int = 1, eta: float = 0.0, 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, output_type: str = "pil", return_dict: bool = True, attention_kwargs: Optional[Dict[str, Any]] = None, callback_on_step_end: Optional[ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks] ] = None, callback_on_step_end_tensor_inputs: List[str] = ["latents"], max_sequence_length: int = 226, controlnet_weights: Optional[Union[float, list, np.ndarray, torch.FloatTensor]] = 1.0, controlnet_guidance_start: float = 0.0, controlnet_guidance_end: float = 1.0, ) -> Union[CogVideoXPipelineOutput, Tuple]: if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)): callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs height = height or self.transformer.config.sample_height * self.vae_scale_factor_spatial width = width or self.transformer.config.sample_width * self.vae_scale_factor_spatial num_frames = num_frames or self.transformer.config.sample_frames num_videos_per_prompt = 1 # 1. Check inputs. Raise error if not correct self.check_inputs( image=image, prompt=prompt, height=height, width=width, negative_prompt=negative_prompt, callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, ) self._guidance_scale = guidance_scale self._attention_kwargs = attention_kwargs self._interrupt = False # 2. Default 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 # 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. do_classifier_free_guidance = guidance_scale > 1.0 # 3. Encode input prompt prompt_embeds, negative_prompt_embeds = self.encode_prompt( prompt=prompt, negative_prompt=negative_prompt, do_classifier_free_guidance=do_classifier_free_guidance, num_videos_per_prompt=num_videos_per_prompt, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, max_sequence_length=max_sequence_length, device=device, ) if do_classifier_free_guidance: prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) # 4. Prepare timesteps timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps) self._num_timesteps = len(timesteps) # 5. Prepare latents latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1 # For CogVideoX 1.5, the latent frames should be padded to make it divisible by patch_size_t patch_size_t = self.transformer.config.patch_size_t additional_frames = 0 if patch_size_t is not None and latent_frames % patch_size_t != 0: additional_frames = patch_size_t - latent_frames % patch_size_t num_frames += additional_frames * self.vae_scale_factor_temporal image = self.video_processor.preprocess(image, height=height, width=width).to( device, dtype=prompt_embeds.dtype ) latent_channels = self.transformer.config.in_channels // 2 latents, image_latents = self.prepare_latents( image, batch_size * num_videos_per_prompt, latent_channels, num_frames, height, width, prompt_embeds.dtype, device, generator, latents, ) # 6 encoding anchor videos anchor_states = self.encode_video(anchor_video[None]).to(device) # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta) # 8. Create rotary embeds if required image_rotary_emb = ( self._prepare_rotary_positional_embeddings(height, width, latents.size(1), device) if self.transformer.config.use_rotary_positional_embeddings else None ) # 8. Create ofs embeds if required ofs_emb = None if self.transformer.config.ofs_embed_dim is None else latents.new_full((1,), fill_value=2.0) # 9. Denoising loop num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0) if do_classifier_free_guidance: anchor_states = torch.cat([anchor_states] * 2) with self.progress_bar(total=num_inference_steps) as progress_bar: # for DPM-solver++ old_pred_original_sample = None for i, t in enumerate(timesteps): if self.interrupt: continue latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) latent_image_input = torch.cat([image_latents] * 2) if do_classifier_free_guidance else image_latents latent_and_image_input = torch.cat([latent_model_input, latent_image_input], dim=2) # latent_and_image_input = torch.cat([latent_model_input], dim=2) # broadcast to batch dimension in a way that's compatible with ONNX/Core ML timestep = t.expand(latent_model_input.shape[0]) current_sampling_percent = i / len(timesteps) latent_model_input = latent_model_input.to(dtype=self.transformer.dtype) prompt_embeds = prompt_embeds.to(dtype=self.transformer.dtype) controlnet_states = None input_controlnet_states = anchor_states if (controlnet_guidance_start <= current_sampling_percent <= controlnet_guidance_end): controlnet_states = self.controlnet( hidden_states=latent_model_input, encoder_hidden_states=prompt_embeds, image_rotary_emb=image_rotary_emb, controlnet_states=input_controlnet_states, controlnet_output_mask = controlnet_output_mask, timestep=timestep, return_dict=False, )[0] if isinstance(controlnet_states, (tuple, list)): controlnet_states = [x.to(dtype=self.transformer.dtype) for x in controlnet_states] else: controlnet_states = controlnet_states.to(dtype=self.transformer.dtype) # predict noise model_output noise_pred = self.transformer( hidden_states=latent_and_image_input, encoder_hidden_states=prompt_embeds, timestep=timestep, image_rotary_emb=image_rotary_emb, controlnet_states=controlnet_states, controlnet_weights=controlnet_weights, return_dict=False, )[0] noise_pred = noise_pred.float() # perform guidance if use_dynamic_cfg: self._guidance_scale = 1 + guidance_scale * ( (1 - math.cos(math.pi * ((num_inference_steps - t.item()) / num_inference_steps) ** 5.0)) / 2 ) if 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 if not isinstance(self.scheduler, CogVideoXDPMScheduler): latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] else: latents, old_pred_original_sample = self.scheduler.step( noise_pred, old_pred_original_sample, t, timesteps[i - 1] if i > 0 else None, latents, **extra_step_kwargs, return_dict=False, ) latents = latents.to(prompt_embeds.dtype) # call the callback, if provided 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) if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): progress_bar.update() if not output_type == "latent": # Discard any padding frames that were added for CogVideoX 1.5 latents = latents[:, additional_frames:] video = self.decode_latents(latents) video = self.video_processor.postprocess_video(video=video, output_type=output_type) else: video = latents # Offload all models self.maybe_free_model_hooks() if not return_dict: return (video,) return CogVideoXPipelineOutput(frames=video)