Nvidia Releases Open-Source SANA-WM, 2.6 Billion-Parameter World Model for Single-GPU 720p Video
Nvidia released SANA-WM, an open-source 2.6 billion-parameter world model for minute-scale video generation that can create 720p videos up to 60 seconds long on a single GPU. Using one image, a text prompt and a six-degrees-of-freedom camera trajectory, the model is designed to produce action-controllable scenes with precise camera movement.
SANA-WM uses hybrid linear attention, including a Hybrid Linear Diffusion Transformer with Gated DeltaNet, to make longer-context video generation more memory-efficient.
From the sources (3 posts)
@brianroemmeleNVIDIA just unleashed SANA-WM and it’s an absolute MONSTER for the future of open source AI! A blazing-fast 2.6B-parameter open-source world model that doesn’t just generate video… it creates controllable, physics-rich, high-fidelity world
@huggingpapersNVIDIA released SANA-WM: Efficient Minute-Scale World Modeling A 2.6B parameter open-source world model generating 720p, minute-long videos with precise 6-DoF camera control on a single GPU. Uses Hybrid Linear Diffusion Transformer with Ga
@xieenze_jrPlease check our SANA-WM, an efficient minute-scale world model with hybrid linear attention! 🎉