face-preservation

ID-V2V is a video-to-video generative framework for identity-preserving video restylization. It treats identity preservation as a video relighting problem and uses edited keyframes for style propagation, achieving high-quality results without paired training data.