Skip to content

Change VAE model source in inference.py#364

Open
zouzhekang wants to merge 1 commit into
bytedance:mainfrom
zouzhekang:zouzhekang-patch-1
Open

Change VAE model source in inference.py#364
zouzhekang wants to merge 1 commit into
bytedance:mainfrom
zouzhekang:zouzhekang-patch-1

Conversation

@zouzhekang

Copy link
Copy Markdown

I fine-tuned the decoder part of the VAE to enable it to produce clearer teeth and lips.

我优化了vae的解码部分,能增强牙齿和嘴唇边缘的清晰度。

I fine-tuned the decoder part of the VAE to enable it to produce clearer teeth and lips.
@zouzhekang

zouzhekang commented May 19, 2026

Copy link
Copy Markdown
Author

主要是制作数据重训sd-vae-ft-mse的解码器部分,对齐方式和latentsync保持一致,原来用MSE会出现更平滑的结果,我改为CharbonnierLoss,另外对嘴部区域的loss做了局部感知加强,实测最终效果无色差和连续性问题,效果如下(仔细看牙齿和嘴唇边缘会清晰一点):
originenhanced

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant