From the repo, it looks like a research release for an interactive world/video generation model. It includes inference code and a 14B causal-fast checkpoint.
A few caveats: the license is non-commercial, and the example setup looks GPU-heavy, so I read this more as a research/demo release than a product-ready system.