World Labs Atlas
A World Labs model that generates, reconstructs, and simulates three dimensional scenes from a handful of photographs.
Announced in September 2026, Atlas takes text, images, video, or 3D data and anchors all of it to positions in space rather than treating it as a flat sequence. In practice that means you can hand it two or three photos of a place and get back views from angles that were never photographed, plus the scene as real 3D data you can load into other software.
It outputs up to a minute of video at 1440p, exports point clouds and Gaussian splats, and can generate what a robot’s cameras would see along a path through a reconstructed room. That last use is the commercially interesting one, because it turns a single real recording into thousands of training situations. Not to be confused with Atlas, the OpenAI web browser that shut down in 2026.