NVIDIA's SANA-WM: 2.6B open-source world model generates minute-long 720p video

NVIDIA Labs has released SANA-WM, a 2.6-billion-parameter open-source world model capable of generating coherent 720p video sequences up to a minute long. The compact parameter count is notable in a field where comparable generative video systems typically run an order of magnitude larger, suggesting architectural efficiency gains rather than brute-force scaling.

Releasing the weights and code openly puts SANA-WM in direct contrast with the closed, API-gated approach taken by most frontier video generation systems. For researchers and independent developers, an openly available model at this capability level lowers the barrier to studying long-horizon temporal consistency, physical plausibility, and the broader question of whether generative models can serve as usable simulators of the world.