LongCat-Video-Avatar
Read the LongCat-Video-Avatar audio-driven avatar video model line, and see how it handles long-clip stability, multi-person identity and natural motion.
LongCat-Video-Avatar, from Meituan’s LongCat team, is an audio-driven avatar video model built on the LongCat-Video architecture. It takes audio and text as conditions, covering audio-text-to-video, image-conditioned video and video continuation for one or many speakers. Cross-chunk latent stitching limits quality loss over long clips, and reference skip attention keeps identity without hard copies of the reference frame.
