SANA-WM: Open-source 2.6B world model generates 1-minute 720p video
Generate full minute-long, high-res videos with a single 2.6B parameter model.
Deep Dive
The article is titled "Comments" and contains no additional information.
Key Points
- 2.6B parameter open-source world model generates 60-second, 720p video at up to 60 FPS
- Uses diffusion transformer with temporal attention for frame-to-frame consistency
- Runs on consumer GPUs (24GB VRAM) with optimizations; weights available on Hugging Face
Why It Matters
Open-source long-form video generation enables researchers and creators to build coherent, high-res simulations without proprietary APIs.