File size: 4,916 Bytes
5b1c701
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
# Copyright 2025 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.

from typing import Any, List, Tuple, Union

import numpy as np
import PIL
import torch

from diffusers.configuration_utils import FrozenDict
from diffusers.models import AutoencoderKLWan
from diffusers.utils import logging
from diffusers.video_processor import VideoProcessor
from diffusers.modular_pipelines import ModularPipelineBlocks, PipelineState
from diffusers.modular_pipelines.modular_pipeline_utils import (
    ComponentSpec,
    InputParam,
    OutputParam,
)
import types


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


class WanRTDecodeStep(ModularPipelineBlocks):
    model_name = "WanRT"
    decoder_cache = []

    @property
    def expected_components(self) -> List[ComponentSpec]:
        return [
            ComponentSpec(
                "vae",
                AutoencoderKLWan,
                repo="Wan-AI/Wan2.1-T2V-14B-Diffusers",
                subfolder="vae",
            ),
            ComponentSpec(
                "video_processor",
                VideoProcessor,
                config=FrozenDict({"vae_scale_factor": 8}),
                default_creation_method="from_config",
            ),
        ]

    @property
    def description(self) -> str:
        return "Step that decodes the denoised latents into images"

    @property
    def inputs(self) -> List[Tuple[str, Any]]:
        return [
            InputParam("output_type", default="pil"),
            InputParam(
                "latents",
                required=True,
                type_hint=torch.Tensor,
                description="The denoised latents from the denoising step",
            ),
            InputParam(
                "frame_cache_context",
                description="The denoised latents from the denoising step",
            ),
            InputParam(
                "block_idx",
                description="The denoised latents from the denoising step",
            ),
            InputParam(
                "decoder_cache",
                description="The denoised latents from the denoising step",
            ),
        ]

    @property
    def intermediate_outputs(self) -> List[str]:
        return [
            OutputParam(
                "videos",
                type_hint=Union[
                    List[List[PIL.Image.Image]], List[torch.Tensor], List[np.ndarray]
                ],
                description="The generated videos, can be a PIL.Image.Image, torch.Tensor or a numpy array",
            )
        ]

    @torch.no_grad()
    def __call__(self, components, state: PipelineState) -> PipelineState:
        block_state = self.get_block_state(state)
        vae_dtype = components.vae.dtype

        # Disable clearing cache
        if block_state.block_idx == 0:
            components.vae.clear_cache()
            components.vae.clear_cache = lambda: None
            components.vae._feat_map = [None] * 55

        if block_state.block_idx != 0:
            components.vae._feat_map = block_state.decoder_cache

        if not block_state.output_type == "latent":
            latents = block_state.latents.to(components.vae.device)

            # Create tensors directly on target device and dtype to avoid redundant conversions
            latents_mean = torch.tensor(
                components.vae.config.latents_mean,
                device=latents.device,
                dtype=latents.dtype,
            ).view(1, components.vae.config.z_dim, 1, 1, 1)
            latents_std = 1.0 / torch.tensor(
                components.vae.config.latents_std,
                device=latents.device,
                dtype=latents.dtype,
            ).view(1, components.vae.config.z_dim, 1, 1, 1)

            latents = latents / latents_std + latents_mean
            latents = latents.to(vae_dtype)

            videos = components.vae.decode(latents, return_dict=False)[0]

        else:
            block_state.videos = block_state.latents

        block_state.decoder_cache = components.vae._feat_map
        block_state.frame_cache_context.extend(videos.split(1, dim=2))

        videos = components.video_processor.postprocess_video(
            videos, output_type=block_state.output_type
        )
        block_state.videos = videos

        self.set_block_state(state, block_state)

        return components, state