World Labs unveils Atlas, a multimodal world model for pixel-level 3D reconstruction

World Labs, co-founded by Fei-Fei Li, unveiled Atlas on September 2, claiming it to be the world's first multimodal world model. The model generates images and video frames with pixel-accurate camera control, reconstructing scenes in 3D. It accepts inputs from text, images, and video, outputting up to 1-minute 1440p videos and 360-degree panoramas.

Atlas's key capabilities include generating novel views and 3D models from one or multiple images, surpassing existing open-source reconstruction models; supporting real-to-sim workflows for robotics through spatiotemporal simulation; and generating images and panoramas from text. Some partners have early access, with broader access opening in the coming weeks.