World Labs releases Atlas, a world model for spatial intelligence
The Fei-Fei Li startup's 'omni' model generates up to a minute of 1440p, camera-controllable video and reconstructs 3D scenes from as few as one to three images; it entered early access with select partners.
- Models & capabilities
- Notable
World Labs, the spatial-intelligence company founded by Fei-Fei Li, released Atlas, which it described as an “omni” world model — a single system operating natively over text, images, video and 3D data rather than a specialist trained for one of them.
The company said Atlas can generate up to a minute of video at 1440p with pixel-level camera control, reconstruct a 3D scene from as few as one to three input images (and in finer detail from a hundred or more), and produce images and 360-degree panoramas from text. On its own figures, Atlas was preferred to Google’s Gemini Omni Flash in 81% of camera-controlled generation comparisons and outperformed open-source specialists at 3D reconstruction despite being a generalist, with quality improving as training compute was scaled. The numbers are World Labs’ own and were not independently verified at release.
Atlas entered early access with a set of selected partners rather than shipping broadly, and no pricing was disclosed. It joins a crowded contest to build usable world models, alongside Google DeepMind’s Genie 3 and Runway’s Solaris — systems that generate navigable, physically plausible environments and that their makers frame as groundwork for robotics and simulation as much as for media.