A video-to-video lip-sync model based on Latent Diffusion, featuring custom code modifications and post-training on 3,000 hours of video data. 

https://github.com/Reopen-AI/-LatentSync_FineTune