World Labs, the startup co-founded by Fei-Fei Li, has unveiled Atlas, a multimodal autoregressive diffusion transformer that generates realistic 3D environments and camera-controlled video from a single 2D image. Atlas can produce up to a minute of 1440p video and reconstructs scenes in 3D with precise camera control, and the company reports it winning 81-93% of head-to-head preference tests against specialized video baselines. Atlas builds on World Labs’ earlier Marble model and is currently available in early access to select enterprises.
