NVIDIA unveils SANA-WM: a model for 60-second 720p video on a single GPU

NVIDIA has unveiled SANA-WM, an open world model for camera-controlled video generation. The model creates 60-second 720p videos with precise 6-DoF control and can run on a single RTX 5090.

How SANA-WM works

SANA-WM is a so-called world model, a simulator of the physical world. Rather than simply stitching images into video, the model learns how the physical world reacts to actions: if you pan the camera to the left, objects in the frame move correctly. The model has 2.6 billion parameters and was trained on 64 H100 GPUs.

Key features of SANA-WM:

• Generation of 60-second videos in 720p (1280×720)
• Precise camera control via 6-DoF (position and orientation)
• Runs on a single RTX 5090 without cloud services
• Open source code for experimentation and adaptation
• Scaling support: from academic research to production

Why it matters for video creators

Before SANA-WM, video generators were either cloud services (expensive) or required specialized hardware. SANA-WM changes that: it runs locally, quickly and without subscriptions. A studio can generate video scenarios, frame-by-frame visualizations and drafts in minutes, with no dependence on the cloud.

For directors and animators, this means a way to quickly test an idea for a visual plan. For 3D artists, it offers a way to automate camera transitions in complex scenes. For marketers, it means quickly generating a promo clip with the right motion dynamics and viewing angles.

What it means

SANA-WM signals a shift from cloud-based video generators to local tools. Just as GPUs once made 3D rendering available on every computer, world models are beginning to do the same for video generation. For the industry, this speeds up the adoption of AI in creative workflows, not because models suddenly got much smarter, but because they now run everywhere.