LongCat-Video-Avatar 1.6
Technique Report Code 🤗 Hugging Face
An expressive avatar model built upon LongCat-Video
Technique Report Code 🤗 Hugging Face
An expressive avatar model built upon LongCat-Video
Introduce what LongCat-Video-Avatar 1.6 is, how it improves upon the 1.5 release, and how the model combines stable avatar generation with practical inference speed.
Show how LongCat-Video-Avatar 1.6 reduces error accumulation over long talking sequences while preserving facial identity, natural speech-related mouth motion, body structure, hand-object interactions, and temporal consistency.
Explore LongCat-Video-Avatar 1.6 across diverse expressive scenarios, including music, singing, live performance, character interaction, and creative animation.
Compare LongCat-Video-Avatar 1.5 and 1.6 under matched inputs, highlighting more natural speech-related mouth movements and reduced error accumulation during long-video generation.
Compare LongCat-Video-Avatar 1.6 with OmniHuman-2.0, HeyGen, Kling Avatar 2.0, and OmniHuman-1.5 under the same inputs.
Part of images and audios are derived from real videos solely to demonstrate the capabilities of this research, e.g., expressions, gestures, and naturalness. The generated content is for academic use only and commercial use is not permitted. If there are any concerns, please contact us (zhangyong202303@gmail.com) and we will delete it in time.