SANA-WM, a 2.6B open-source world model for 1-minute 720p video

SANA-WM is a 2.6B-parameter open-source world model that can generate 1-minute 720p video, setting a new bar for accessible long-form video AI. It outperforms #9's typical proprietary models in both parameter efficiency and output duration, producing coherent scenes with 40% fewer artifacts. The model’s open-source nature allows researchers to fine-tune it without licensing costs, a stark contrast to closed alternatives that charge $0.05 per second of generated video. This capability could democratize video creation, as SANA-WM’s inference speed is 2.3 times faster than the average commercial competitor, enabling real-time editing workflows.