Meituan Releases LongCat-Video-Avatar 1.6 for Audio-Driven Video Generation
Meituan has released LongCat-Video-Avatar 1.6, an updated open-source framework featuring improved lip synchronization, stable long-form video generation exceeding 10 minutes, and faster inference optimizations. This release significantly advances open-source audio-driven human video generation by solving common issues like temporal instability and error accumulation in long videos. It empowers developers and creators to build realistic digital avatars and long-form conversational video applications more efficiently. Version 1.6 minimizes exaggerated or strained mouth movements, mitigates error accumulation during autoregressive continuation, and introduces pipeline optimizations that accelerate end-to-end inference execution.
## BACKGROUND
Audio-driven human video generation involves creating realistic footage of a person speaking and moving in sync with a given audio track. Autoregressive video continuation is a technique where a model generates video content sequentially by predicting future frames based on previously generated history.