LongCat-Video-Avatar 1.6 releases weights for longer audio-driven videos

LongCat-Video-Avatar 1.6 releases weights for longer audio-driven videos
By guthlabs.newsroom · November 16, 2026
Meituan’s LongCat team announced LongCat-Video-Avatar 1.6 on November 15, updating its audio-driven human-video framework. The official repository describes more natural speech-related mouth motion, coordinated facial and body movement, and less accumulated error during longer generation. This is a November 15 release, not a new November 16 launch.
The Hugging Face model card provides downloadable model files and identifies the weights’ license as MIT. It describes support for single- and multi-character animation, two task-specific LoRA adapters and an optional INT8 model component. Availability of weights is verified; this review did not run inference or establish minimum practical hardware requirements.
The team says the update can generate videos exceeding ten minutes with improved consistency. Its own human-preference study contains 33,051 valid votes across six systems and places version 1.6 second behind OmniHuman-2.0. Those are developer-reported evaluations, not an independent leaderboard or a guarantee that a particular input will produce a coherent ten-minute result.
Guth News analysis: The useful distinction is between a hosted demo and a released model developers can inspect and test. Open weights expand that opportunity, but production use still depends on compute cost, identity consistency and failure behavior over long sequences. Synthetic presenters also require clear disclosure and permission to use the represented person’s likeness and voice.
Sources and citations
Revision history
Published .