LongCat-Avatar

TL;DR

LongCat-Video-Avatar is an audio-driven character-animation model in Meituan's LongCat-Video project. The recovered post points to a real repository, but its claim of free, minutes-long talking videos is not a local performance or operating-cost result. Official documentation supports audio/image-driven generation and video continuation; no generation was tested during this review.

What it actually is

Primary sources checked on 2026-09-27:

• The official repository documents Avatar 1.0 and 1.5, with audio/image-to-video and continuation examples. The capture does not establish which version the post demonstrated. Repository README
• The repository's software license is MIT; the official Avatar 1.5 model card separately states that its weights use MIT. This does not establish a zero-cost hosted service. Code license, 1.5 model card
• Installation instructions use Python, PyTorch/CUDA, and model downloads. Avatar 1.5 examples use multi-process GPU execution and offer INT8 loading to reduce VRAM. These examples are not a verified minimum hardware requirement; no particular GPU, memory budget, runtime, or cloud price is asserted here. README

Links

• Repo: https://github.com/meituan-longcat/LongCat-Video
• Docs: https://meigen-ai.github.io/LongCat-Video-Avatar-1.5-Page/
• Model card: https://huggingface.co/meituan-longcat/LongCat-Video-Avatar-1.5
• Original source: https://www.instagram.com/p/DcWKE-cAcy7/

Kickstarter guide

Read the current Avatar-specific README and model card before selecting a version. Assess compatible compute and expected operating cost before downloading weights or running inference. If a future trial is warranted, start with one short narration and a portrait you have permission to animate; record setup effort, runtime, lip synchronization, visual consistency, and failures. Do not treat the project's demonstration media as reusable portfolio assets: its project page limits those research demonstrations to academic use. Project page