Creators and spatial-intelligence developers can explore persistent worlds rather than generating each shot as an isolated clip. Core capabilities include Multi-turn natural-language direction, Interleaved camera and text control, Up to 60-second sequences, and 720p video output.
Public code or model weights are linked from the project, with setup instructions for running or adapting the system. Additional characteristics include 360K-token context, Persistent scene memory, Multiple visual styles, and First-person and third-person viewpoints.
