Efficient-Large-Model/SANA-WM_bidirectional
Researchers have released SANA-WM, an open-source world model capable of generating minute-long videos at 720p resolution. This diffusion transformer model utilizes a hybrid linear attention mechanism and a dual-branch architecture for precise camera control. The model also incorporates a two-stage generation pipeline with a refiner for enhanced quality and temporal consistency, and it was trained using a robust annotation pipeline with metric-scale 6-DoF camera poses. AI
IMPACT Enables creation of longer, high-fidelity videos with precise camera control, potentially impacting content generation and simulation.