mirror of
https://github.com/modelscope/DiffSynth-Studio.git
synced 2026-03-18 13:58:15 +00:00
44 lines
2.3 KiB
Python
44 lines
2.3 KiB
Python
import torch
|
||
from PIL import Image
|
||
from diffsynth.utils.data import save_video, VideoData
|
||
from diffsynth.core import load_state_dict
|
||
from diffsynth.pipelines.wan_video import WanVideoPipeline, ModelConfig
|
||
|
||
|
||
vram_config = {
|
||
"offload_dtype": torch.bfloat16,
|
||
"offload_device": "cpu",
|
||
"onload_dtype": torch.bfloat16,
|
||
"onload_device": "cpu",
|
||
"preparing_dtype": torch.bfloat16,
|
||
"preparing_device": "cuda",
|
||
"computation_dtype": torch.bfloat16,
|
||
"computation_device": "cuda",
|
||
}
|
||
pipe = WanVideoPipeline.from_pretrained(
|
||
torch_dtype=torch.bfloat16,
|
||
device="cuda",
|
||
model_configs=[
|
||
ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="high_noise_model/diffusion_pytorch_model*.safetensors"),
|
||
ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="low_noise_model/diffusion_pytorch_model*.safetensors"),
|
||
ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth", **vram_config),
|
||
ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="Wan2.1_VAE.pth", **vram_config),
|
||
],
|
||
)
|
||
state_dict = load_state_dict("models/train/Wan2.2-VACE-Fun-A14B_high_noise_full/epoch-1.safetensors", torch_dtype=torch.bfloat16, device="cpu")
|
||
pipe.vace.load_state_dict(state_dict)
|
||
state_dict = load_state_dict("models/train/Wan2.2-VACE-Fun-A14B_low_noise_full/epoch-1.safetensors", torch_dtype=torch.bfloat16, device="cpu")
|
||
pipe.vace2.load_state_dict(state_dict)
|
||
|
||
video = VideoData("data/example_video_dataset/video1_softedge.mp4", height=480, width=832)
|
||
video = [video[i] for i in range(17)]
|
||
reference_image = VideoData("data/example_video_dataset/video1.mp4", height=480, width=832)[0]
|
||
|
||
video = pipe(
|
||
prompt="from sunset to night, a small town, light, house, river",
|
||
negative_prompt="色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走",
|
||
vace_video=video, vace_reference_image=reference_image, num_frames=17,
|
||
seed=1, tiled=True
|
||
)
|
||
save_video(video, "video_Wan2.2-VACE-A14B.mp4", fps=15, quality=5)
|