DiffSynth-Studio/examples/wanvideo/model_inference/Wan2.2-Animate-14B.py at main · IIIcecola/DiffSynth-Studio · GitHub

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
import torch
from PIL import Image
from diffsynth import save_video, VideoData, load_state_dict
from diffsynth.pipelines.wan_video_new import WanVideoPipeline, ModelConfig
from modelscope import dataset_snapshot_download, snapshot_download


pipe = WanVideoPipeline.from_pretrained(
    torch_dtype=torch.bfloat16,
    device="cuda",
    model_configs=[
        ModelConfig(model_id="Wan-AI/Wan2.2-Animate-14B", origin_file_pattern="diffusion_pytorch_model*.safetensors", offload_device="cpu"),
        ModelConfig(model_id="Wan-AI/Wan2.2-Animate-14B", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth", offload_device="cpu"),
        ModelConfig(model_id="Wan-AI/Wan2.2-Animate-14B", origin_file_pattern="Wan2.1_VAE.pth", offload_device="cpu"),
        ModelConfig(model_id="Wan-AI/Wan2.2-Animate-14B", origin_file_pattern="models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth", offload_device="cpu"),
    ],
)
pipe.enable_vram_management()

dataset_snapshot_download(
    dataset_id="DiffSynth-Studio/examples_in_diffsynth",
    local_dir="./",
    allow_file_pattern="data/examples/wan/animate/*",
)

# Animate
input_image = Image.open("data/examples/wan/animate/animate_input_image.png")
animate_pose_video = VideoData("data/examples/wan/animate/animate_pose_video.mp4").raw_data()[:81-4]
animate_face_video = VideoData("data/examples/wan/animate/animate_face_video.mp4").raw_data()[:81-4]
video = pipe(
    prompt="视频中的人在做动作",
    seed=0, tiled=True,
    input_image=input_image,
    animate_pose_video=animate_pose_video,
    animate_face_video=animate_face_video,
    num_frames=81, height=720, width=1280,
    num_inference_steps=20, cfg_scale=1,
)
save_video(video, "video1.mp4", fps=15, quality=5)

# Replace
snapshot_download("Wan-AI/Wan2.2-Animate-14B", allow_file_pattern="relighting_lora.ckpt", local_dir="models/Wan-AI/Wan2.2-Animate-14B")
lora_state_dict = load_state_dict("models/Wan-AI/Wan2.2-Animate-14B/relighting_lora.ckpt", torch_dtype=torch.float32, device="cuda")["state_dict"]
pipe.load_lora(pipe.dit, state_dict=lora_state_dict)
input_image = Image.open("data/examples/wan/animate/replace_input_image.png")
animate_pose_video = VideoData("data/examples/wan/animate/replace_pose_video.mp4").raw_data()[:81-4]
animate_face_video = VideoData("data/examples/wan/animate/replace_face_video.mp4").raw_data()[:81-4]
animate_inpaint_video = VideoData("data/examples/wan/animate/replace_inpaint_video.mp4").raw_data()[:81-4]
animate_mask_video = VideoData("data/examples/wan/animate/replace_mask_video.mp4").raw_data()[:81-4]
video = pipe(
    prompt="视频中的人在做动作",
    seed=0, tiled=True,
    input_image=input_image,
    animate_pose_video=animate_pose_video,
    animate_face_video=animate_face_video,
    animate_inpaint_video=animate_inpaint_video,
    animate_mask_video=animate_mask_video,
    num_frames=81, height=720, width=1280,
    num_inference_steps=20, cfg_scale=1,
)
save_video(video, "video2.mp4", fps=15, quality=5)