telestylevideo_pipeline.py

# Copyright 2025 The Wan Team and The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import html
from typing import Any, Callable, Dict, List, Optional, Union

import ftfy
import regex as re
import torch
from transformers import AutoTokenizer, UMT5EncoderModel

from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback
from diffusers.loaders import WanLoraLoaderMixin
from diffusers.models import AutoencoderKLWan
from telestylevideo_transformer import WanTransformer3DModel
from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
from diffusers.utils import is_torch_xla_available, logging, replace_example_docstring
from diffusers.utils.torch_utils import randn_tensor
from diffusers.video_processor import VideoProcessor
from diffusers.pipelines.pipeline_utils import DiffusionPipeline
from diffusers.pipelines.wan.pipeline_output import WanPipelineOutput


if is_torch_xla_available():
    import torch_xla.core.xla_model as xm

    XLA_AVAILABLE = True
else:
    XLA_AVAILABLE = False

logger = logging.get_logger(__name__)  # pylint: disable=invalid-name


EXAMPLE_DOC_STRING = """
    Examples:
        ```python
        >>> import torch
        >>> from diffusers.utils import export_to_video
        >>> from diffusers import AutoencoderKLWan, WanPipeline
        >>> from diffusers.schedulers.scheduling_unipc_multistep import UniPCMultistepScheduler

        >>> # Available models: Wan-AI/Wan2.1-T2V-14B-Diffusers, Wan-AI/Wan2.1-T2V-1.3B-Diffusers
        >>> model_id = "Wan-AI/Wan2.1-T2V-14B-Diffusers"
        >>> vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
        >>> pipe = WanPipeline.from_pretrained(model_id, vae=vae, torch_dtype=torch.bfloat16)
        >>> flow_shift = 5.0  # 5.0 for 720P, 3.0 for 480P
        >>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config, flow_shift=flow_shift)
        >>> pipe.to("cuda")

        >>> prompt = "A cat and a dog baking a cake together in a kitchen. The cat is carefully measuring flour, while the dog is stirring the batter with a wooden spoon. The kitchen is cozy, with sunlight streaming through the window."
        >>> negative_prompt = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards"

        >>> output = pipe(
        ...     prompt=prompt,
        ...     negative_prompt=negative_prompt,
        ...     height=720,
        ...     width=1280,
        ...     num_frames=81,
        ...     guidance_scale=5.0,
        ... ).frames[0]
        >>> export_to_video(output, "output.mp4", fps=16)
        ```
"""


def basic_clean(text):
    text = ftfy.fix_text(text)
    text = html.unescape(html.unescape(text))
    return text.strip()


def whitespace_clean(text):
    text = re.sub(r"\s+", " ", text)
    text = text.strip()
    return text


def prompt_clean(text):
    text = whitespace_clean(basic_clean(text))
    return text


class WanPipeline(DiffusionPipeline, WanLoraLoaderMixin):
    r"""
    Pipeline for text-to-video generation using Wan.

    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
    implemented for all pipelines (downloading, saving, running on a particular device, etc.).

    Args:
        tokenizer ([`T5Tokenizer`]):
            Tokenizer from [T5](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5Tokenizer),
            specifically the [google/umt5-xxl](https://huggingface.co/google/umt5-xxl) variant.
        text_encoder ([`T5EncoderModel`]):
            [T5](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5EncoderModel), specifically
            the [google/umt5-xxl](https://huggingface.co/google/umt5-xxl) variant.
        transformer ([`WanTransformer3DModel`]):
            Conditional Transformer to denoise the input latents.
        scheduler ([`UniPCMultistepScheduler`]):
            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
        vae ([`AutoencoderKLWan`]):
            Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
    """

    model_cpu_offload_seq = "text_encoder->transformer->vae"
    _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]

    def __init__(
        self,
        #tokenizer: AutoTokenizer,
        #text_encoder: UMT5EncoderModel,
        transformer: WanTransformer3DModel,
        vae: AutoencoderKLWan,
        scheduler: FlowMatchEulerDiscreteScheduler,
    ):
        super().__init__()
        self.register_modules(
            vae=vae,
            transformer=transformer,
            scheduler=scheduler,
        )

        # self.register_modules(
        #     vae=vae,
        #     text_encoder=text_encoder,
        #     tokenizer=tokenizer,
        #     transformer=transformer,
        #     scheduler=scheduler,
        # )

        self.vae_scale_factor_temporal = 2 ** sum(self.vae.temperal_downsample) if getattr(self, "vae", None) else 4
        self.vae_scale_factor_spatial = 2 ** len(self.vae.temperal_downsample) if getattr(self, "vae", None) else 8
        self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)

    # 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
    #     prompt = [prompt_clean(u) for u in 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_attention_mask=True,
    #         return_tensors="pt",
    #     )
    #     text_input_ids, mask = text_inputs.input_ids, text_inputs.attention_mask
    #     seq_lens = mask.gt(0).sum(dim=1).long()

    #     prompt_embeds = self.text_encoder(text_input_ids.to(device), mask.to(device)).last_hidden_state
    #     prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
    #     prompt_embeds = [u[:v] for u, v in zip(prompt_embeds, seq_lens)]
    #     prompt_embeds = torch.stack(
    #         [torch.cat([u, u.new_zeros(max_sequence_length - u.size(0), u.size(1))]) for u in prompt_embeds], dim=0
    #     )

    #     # 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

    # 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 check_inputs(
    #     self,
    #     prompt,
    #     negative_prompt,
    #     height,
    #     width,
    #     prompt_embeds=None,
    #     negative_prompt_embeds=None,
    #     callback_on_step_end_tensor_inputs=None,
    # ):
    #     if height % 16 != 0 or width % 16 != 0:
    #         raise ValueError(f"`height` and `width` have to be divisible by 16 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 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`: {negative_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 negative_prompt is not None and (
    #         not isinstance(negative_prompt, str) and not isinstance(negative_prompt, list)
    #     ):
    #         raise ValueError(f"`negative_prompt` has to be of type `str` or `list` but is {type(negative_prompt)}")

    def prepare_latents(
        self,
        batch_size: int,
        num_channels_latents: int = 16,
        height: int = 480,
        width: int = 832,
        num_frames: int = 81,
        dtype: Optional[torch.dtype] = None,
        device: Optional[torch.device] = None,
        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
        latents: Optional[torch.Tensor] = None,
    ) -> torch.Tensor:
        if latents is not None:
            return latents.to(device=device, dtype=dtype)

        num_latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
        shape = (
            batch_size,
            num_channels_latents,
            num_latent_frames,
            int(height) // self.vae_scale_factor_spatial,
            int(width) // self.vae_scale_factor_spatial,
        )
        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 do_classifier_free_guidance(self):
        return self._guidance_scale > 1.0

    @property
    def num_timesteps(self):
        return self._num_timesteps

    @property
    def current_timestep(self):
        return self._current_timestep

    # @property
    # def interrupt(self):
    #     return self._interrupt

    # @property
    # def attention_kwargs(self):
    #     return self._attention_kwargs

    @torch.no_grad()
    #@replace_example_docstring(EXAMPLE_DOC_STRING)
    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        negative_prompt: Union[str, List[str]] = None,
        height: int = 480,
        width: int = 832,
        num_frames: int = 81,
        num_inference_steps: int = 50,
        guidance_scale: float = 5.0,
        num_videos_per_prompt: Optional[int] = 1,
        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
        latents: Optional[torch.Tensor] = None,
        prompt_embeds: Optional[torch.Tensor] = None,
        prompt_embeds_: Optional[torch.Tensor] = None,
        negative_prompt_embeds: Optional[torch.Tensor] = None,
        output_type: Optional[str] = "np",
        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 = 512,
        source_latents: Optional[torch.Tensor] = None,
        first_latents: Optional[torch.Tensor] = None,
        neg_first_latents: Optional[torch.Tensor] = None,
    ):

        if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
            callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs

        # 1. Check inputs. Raise error if not correct
        # self.check_inputs(
        #     prompt,
        #     negative_prompt,
        #     height,
        #     width,
        #     prompt_embeds,
        #     negative_prompt_embeds,
        #     callback_on_step_end_tensor_inputs,
        # )

        # if num_frames % self.vae_scale_factor_temporal != 1:
        #     logger.warning(
        #         f"`num_frames - 1` has to be divisible by {self.vae_scale_factor_temporal}. Rounding to the nearest number."
        #     )
        #     num_frames = num_frames // self.vae_scale_factor_temporal * self.vae_scale_factor_temporal + 1
        # num_frames = max(num_frames, 1)

        self._guidance_scale = guidance_scale
        # self._attention_kwargs = attention_kwargs
        self._current_timestep = None
        self._interrupt = False

        device = self._execution_device

        # 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]

        # batch_size = 1

        # 3. Encode input prompt
        # prompt_embeds, negative_prompt_embeds = self.encode_prompt(
        #     prompt=prompt,
        #     negative_prompt=negative_prompt,
        #     do_classifier_free_guidance=self.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,
        # )

        transformer_dtype = self.transformer.dtype
        # prompt_embeds = prompt_embeds.to(transformer_dtype)
        # if negative_prompt_embeds is not None:
        #     negative_prompt_embeds = negative_prompt_embeds.to(transformer_dtype)

        # 4. Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps, device=device)
        timesteps = self.scheduler.timesteps

        # 5. Prepare latent variables
        # num_channels_latents = self.transformer.config.in_channels
        # latents = self.prepare_latents(
        #     batch_size * num_videos_per_prompt,
        #     num_channels_latents,
        #     height,
        #     width,
        #     num_frames,
        #     torch.float32,
        #     device,
        #     generator,
        #     latents,
        # )

        latents = torch.randn_like(source_latents)

        # 6. Denoising loop
        # num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
        self._num_timesteps = len(timesteps)

        condition_latent_model_input = first_latents
        neg_condition_latent_model_input = neg_first_latents
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):

                self._current_timestep = t
                latent_model_input = torch.cat([source_latents, latents.to(transformer_dtype)], dim=1)
                timestep = t.expand(latents.shape[0])
                
                print(timestep, torch.zeros_like(timestep))
                noise_pred = self.transformer(
                    condition_hidden_states=condition_latent_model_input,
                    hidden_states=latent_model_input,
                    condition_timestep=torch.zeros_like(timestep),
                    timestep=timestep,
                    encoder_hidden_states=prompt_embeds_,
                    return_dict=False,
                )[0]

                if self.do_classifier_free_guidance:
                    noise_uncond = self.transformer(
                        condition_hidden_states=neg_condition_latent_model_input,
                        hidden_states=latent_model_input,
                        condition_timestep=torch.zeros_like(timestep),
                        timestep=timestep,
                        encoder_hidden_states=prompt_embeds_,
                        return_dict=False,
                    )[0]
                    noise_pred = noise_uncond + guidance_scale * (noise_pred - noise_uncond)

                # compute the previous noisy sample x_t -> x_t-1
                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]

                # if callback_on_step_end is not None:
                #     callback_kwargs = {}
                #     for k in callback_on_step_end_tensor_inputs:
                #         callback_kwargs[k] = locals()[k]
                #     callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                #     latents = callback_outputs.pop("latents", latents)
                #     prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
                #     negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)

                # call the callback, if provided
                #if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                progress_bar.update()

                # if XLA_AVAILABLE:
                #     xm.mark_step()

        self._current_timestep = None

        if not output_type == "latent":
            latents = latents.to(self.vae.dtype)
            latents_mean = (
                torch.tensor(self.vae.config.latents_mean)
                .view(1, self.vae.config.z_dim, 1, 1, 1)
                .to(latents.device, latents.dtype)
            )
            latents_std = 1.0 / torch.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
                latents.device, latents.dtype
            )
            latents = latents / latents_std + latents_mean
            video = self.vae.decode(latents, return_dict=False)[0]
            video = self.video_processor.postprocess_video(video, output_type=output_type)
        else:
            video = latents

        # Offload all models
        # self.maybe_free_model_hooks()

        if not return_dict:
            return (video,)

        return WanPipelineOutput(frames=video)