Linum v2 is a pair of text-to-video models that turn short text prompts into clips. You can download and use or tweak them since the weights are fully open.
Each one runs on about 2 billion parameters. They’re built for video generation and support clips from 2 to 5 seconds long. You get 360p or 720p quality.
They come under Apache 2.0, so people can use them in research or even commercial tools without tight rules. The weights live on Hugging Face.
These models were built from the ground up by two brothers, Sahil and Manu Chopra, after 2 years of work at their AI lab, Linum. They didn’t reuse old image models. They trained it on video directly, with custom design.
A few things stand out.
Open weights and open license. Most high-end models stay closed. Linum v2 is different.
Built fully on video data. It didn’t start from image models like others often do.
They’re open about strengths and issues. Their blog breaks it all down.
The models are drawing attention on Hacker News and LinkedIn, mostly for being free to use and easy to play with. You can run them locally if your GPU has enough VRAM. Higher res and longer clips aren’t there yet, but might come later.
If you'd like to access this model, you can explore the following possibilities:
Weights