Video-Retalking

VideoReTalking edits the lip movements in an existing talking-head video so they match a new audio track, which suits dubbing, translation, and post-production fixes. It works on real-world footage through a pipeline that aligns expression, generates a synced mouth region, and refines the result with pretrained models. A Colab notebook and Replicate demo are provided. Developed by researchers at Xidian University, Tencent AI Lab, and Tsinghua University.