LongCat-Video-Avatar 1.6 Open Weights Released
Posted As of
The Meituan LongCat team released LongCat-Video-Avatar 1.6, an open-source framework for audio-driven human video generation, on 2026-11-15. The weights are on Hugging Face, and the release emphasizes improved lip movement and reduced error accumulation in long videos.
What happened
Source: Meituan LongCat team's GitHub repository and Hugging Face model card
- On 2026-11-15, LongCat-Video-Avatar 1.6 was released, introduced as an open-source framework for audio-driven human video generation (official)
- Improvements cited include more natural lip movement, more expressive facial and head/body motion, and reduced error accumulation when generating long videos (official)
- The checkpoint uses the same structure as Avatar 1.5 and adds 2 LoRAs (official)
- The weights are published on Hugging Face (official)
- Speed-up figures and per-clip processing time are not confirmed on the official page
Our take
This is an open-weight model that PC users with a GPU can download and run themselves, so it matters for local AI users. However, the required VRAM, runtime, and Mac support have not been confirmed. A secondary article says the license is MIT, but we could not verify this on the model card. Reports that the default LoRA blending settings differ between the model card and README have also not been sorted out yet.
Sources and related posts
- LongCat-Video GitHub repository Primary source
- LongCat-Video-Avatar-1.6 Hugging Face model card Primary source