Wan超快视频加速模型Self Forcing图生视频
5.0
0 reviewsDescription
项目:Wan2.1-T2V-1.3B-Self-Forcing-VACE 5秒视频仅仅需要48秒,
仙宫云注册:https://www.xiangongyun.com/register/9YM3YN
Pix工作室打开即用:https://www.xiangongyun.com/image/detail/de2afed6-3a25-4fcd-87f8-11f537a322d2
Wan2.1-T2V-1.3B-Self-Forcing-VACE 是基于Wan2.1生态的13亿参数文生视频模型。其核心创新在于应用了Self-Forcing训练策略,有效弥合了自回归视频扩散模型在训练与推理(测试)时的分布差距,显著提升了生成视频的时间连贯性与质量。结合VACE模块,它支持更灵活的输入(如首尾帧图像)。优势在于高效、低延迟(支持实时生成),能在消费级显卡上快速(仅需4-10步)生成高质量视频,速度远超同类模型,经过测试,5秒视频仅仅需要48秒!!!大大缩短了时间提高了视频的可玩性。
Discussion
(No comments yet)
Loading...
Reviews
No reviews yet
Versions (1)
- latest (7 months ago)
Node Details
Primitive Nodes (15)
JWInteger (1)
LoadWanVideoT5TextEncoder (1)
Primitive integer [Crystools] (2)
WanVideoBlockSwap (1)
WanVideoDecode (1)
WanVideoEnhanceAVideo (1)
WanVideoExperimentalArgs (1)
WanVideoModelLoader (1)
WanVideoSLG (1)
WanVideoSampler (1)
WanVideoTextEncode (1)
WanVideoVACEEncode (1)
WanVideoVACEStartToEndFrame (1)
WanVideoVAELoader (1)
Custom Nodes (7)
ComfyUI
- LoadImage (1)
- LayerUtility: ImageScaleByAspectRatio V2 (1)
- VHS_VideoCombine (2)
- ColorMatch (1)
- ImageConcanate (1)
- Text Multiline (1)
Model Details
Checkpoints (0)
LoRAs (0)