LongCat-Video
MIT-licensed video model from Meituan for text, image and long-video generation with continuation.
LongCat-Video covers text-to-video, image-to-video and video continuation.
LongCat-Video is an open video generation model from Meituan that handles text-to-video, image-to-video and video continuation in one model. It suits developers who want MIT-licensed weights for long clips.
What it is
The repo includes inference code for text-to-video, image-to-video, video continuation, long-video generation and interactive generation, plus audio-driven avatar models and a Streamlit interface.
Who it's for
It suits developers with CUDA GPUs and the patience to build FlashAttention. It does not suit non-technical users, and the README states no VRAM figure, so size your hardware by trial.
Key features
Three tasks, one model
Text-to-video, image-to-video and continuation share a model.
Long-video generation
Built to extend clips well beyond a single short generation.
Interactive generation
The README lists an interactive video mode.
Avatar models
Audio-driven single and multi-audio avatar video is included.
Streamlit interface
A bundled Streamlit app avoids writing inference scripts.
Multi-GPU and int8
Multi-GPU runs and an int8 option for the avatar model.
Up and running in 3 steps
Our honest verdict
What's great
- Repository code and the model weights are both MIT-licensed, per the README.
- One model handles text, image and continuation, so you do not juggle separate checkpoints.
- Includes a Streamlit interface in addition to command-line scripts.
- Active: the README lists LongCat-Video-Avatar-1.5 released in May 2026.
Where it falls short
- The README states no minimum VRAM, so you must find your own limits with the large weights.
- Requires CUDA 12.4 PyTorch builds and FlashAttention-2, which can take effort to compile.
- Aimed at developers; there is no installer or desktop app.
- The repo has no tagged releases, so versions are tracked by commit and Hugging Face model pages.
Pricing & license
Free under the MIT licence. The only cost is the GPU hardware and storage you run it on; no paid hosted plan appears in the repo.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Is LongCat-Video free for commercial use?
The README says the model weights are released under the MIT License, which allows commercial use. It adds that this grants no rights to Meituan trademarks or patents.
What can LongCat-Video generate?
Text-to-video, image-to-video, video continuation, long-video generation, interactive video and audio-driven avatar video, per the README.
How much GPU memory does it need?
The README gives no VRAM figure. It mentions multi-GPU runs and an int8 option that cuts memory for the avatar model.
