基于Wan2.2 Animate模型进行角色动作迁移或服装更换的二合一工作流.
请注意: 本工作流仅用于技术研究, 禁止用于非法用途. 下载本工作流视为同意此条款, 因违规修改或使用此工作流产生的的法律纠纷, 本人概不负责.
v2.0: 基于我自己修改的SeC插件, 实现了高速高精度低显存用量的服装跟踪遮罩.
v1.0: 首次上传, 使用了我自己修改的 Segformer 检测插件, 比市面上的大部分遮罩工作流快一倍以上.
A combined workflow for character motion transfer and clothing replacement based on the Wan2.2 Animate model.
Note: This workflow is for research and technical study purposes only. Any illegal use is strictly prohibited. Downloading this workflow implies agreement to these terms. I am not responsible for any legal issues arising from modification or misuse of this workflow.
v2.0: Integrated my customized SeC plugin to achieve high-speed, high-accuracy clothing tracking masks with low VRAM usage.
v1.0: Initial release, uses my customized Segformer detection plugin, which is more than twice as fast as most masking workflows currently available.
Description
v2.0更新
FAQ
Comments (4)
i've got this error, help
* SeCVideoSegmentation 465:
- Value not in list: tracking_direction: '1' not in ['forward', 'backward', 'bidirectional']
- Value -1 smaller than min of 0: annotation_frame_idx
* SeCVideoSegmentation 464:
- Value not in list: tracking_direction: '1' not in ['forward', 'backward', 'bidirectional']
- Value -1 smaller than min of 0: annotation_frame_idx
Output will be ignored
kinesis_14b_t2v_cam_0.9_fixKEY.safetensors. Any link or other file?
感谢大佬，工作流很棒，
就是我用Wan2_2-Animate-14B_fp8_scaled_e4m3fn_KJ_v2.safetensors，
有时会出现报错
[INFO] Loading LoRA: WanAnimate\lightx2v_elite_it2v_animate_face with strength: 0.5
[WARNING] lora key not loaded: face_adapter.fuser_blocks.0.linear2.weight
[WARNING] lora key not loaded: face_adapter.fuser_blocks.1.linear2.weight
[WARNING] lora key not loaded: face_encoder.out_proj.bias
[WARNING] lora key not loaded: face_encoder.out_proj.weight
[WARNING] lora key not loaded: head.head.bias
[WARNING] lora key not loaded: head.head.weight
[WARNING] lora key not loaded: head.modulation
[WARNING] lora key not loaded: img_emb.proj.3.bias
[WARNING] lora key not loaded: img_emb.proj.3.weight
[WARNING] lora key not loaded: patch_embedding.bias
[WARNING] lora key not loaded: patch_embedding.weight
[WARNING] lora key not loaded: text_embedding.0.bias
[WARNING] lora key not loaded: text_embedding.0.weight
[WARNING] lora key not loaded: text_embedding.2.bias
[WARNING] lora key not loaded: text_embedding.2.weight
[WARNING] lora key not loaded: time_embedding.2.bias
[WARNING] lora key not loaded: time_embedding.2.weight
似乎只有换成BF16的Animate模型才能解决
只是BF16实在太大，所以想问问大佬有没有更好的办法？
What is the recommand hardware for this workflow? I am using 5080 laptop and it took me almost an hour to process a 6 seconds video, even after I lower some parameters. I am also using your advanced SeC plugin, but it doesn't help much since my laptop took a long time running "WanVideo Sampler". Do you have any suggestion on speeding this up? (I know VRAM matters)