NVIDIA has released SANA-WM, an open-source world model capable of generating minute-long, 720p videos on a single GPU. This model addresses the computational challenges of creating high-resolution, long-duration video for applications in embodied AI and robotics. SANA-WM utilizes a novel architecture with Gated DeltaNet and hybrid linear attention to manage computational complexity, and it offers multiple inference variants for different use cases, including a distilled version that can produce a 60-second clip in under 35 seconds on an RTX 5090. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Enables high-resolution, long-duration video generation on consumer hardware, potentially accelerating research in robotics and simulation.
RANK_REASON Open-source model release with technical details and architecture description. [lever_c_demoted from research: ic=1 ai=1.0]
