✦
Summary
SANA-WM is NVIDIA's efficient minute-scale world model for generating controllable 720p video from an initial image and a camera trajectory. It is designed to make long-horizon world simulation more practical by producing minute-long videos on a single GPU while preserving camera control. The model targets interactive environments, embodied AI, simulation, and generative video workflows where both quality and temporal length matter.
Key Features
Generates 720p minute-scale video from one image and a camera trajectory.
Uses a 2.6B-parameter Hybrid Linear Diffusion Transformer architecture.
Combines Gated DeltaNet and softmax attention for memory-efficient long-context modeling.
Supports precise 6-DoF camera control through a dual-branch camera-control design.
Applies a two-stage generation pipeline with long-video refinement.
Trains from public video clips with metric-scale camera-pose supervision.
Targets interactive world modeling, embodied AI, and camera-controlled video generation.
Provides public paper, code, and model resources for research use.
About SANA-WM
The architecture is a 2.6B-parameter open-source world model with a Hybrid Linear Diffusion Transformer. It combines frame-wise Gated DeltaNet and softmax attention for long-context modeling, uses dual-branch camera control for 6-DoF trajectory adherence, and applies a two-stage pipeline with a long-video refiner. These design choices help SANA-WM maintain temporal consistency and visual quality over longer sequences than typical short-form video generators.
SANA-WM is valuable for researchers and developers building explorable AI worlds, robotics simulators, camera-controlled video tools, or data engines for embodied agents. Its efficient training and inference profile makes it notable because it uses public video data with metric-scale pose supervision rather than depending only on massive closed datasets. The release provides paper, code, and model links, so it is listed as a free open-source world-model project.