InfiniteTalk

Why we included this project

Most talking-avatar models only manage short clips, but InfiniteTalk keeps the video going as long as the audio track does, which makes it useful for dubbing a full scene, podcast, or lecture in one pass instead of stitching together ten-second chunks. You feed it a single face image or a sparse set of video frames plus a voice track, and it outputs a talking video with lip-synced speech, head motion, and facial expressions that follow the audio. It handles both image-to-video and video-to-video generation, and the code and Hugging Face weights are openly available, so a team can actually run it rather than just read the paper. The same group has since released a newer avatar framework aimed at more production-ready human video, which makes InfiniteTalk a reasonable starting point if you are exploring audio-driven avatars.