World Labs Unveils Atlas, a World Model for Video, Images, and 3D Space
Atlas combines video, image, and 3D space generation in one model, and human evaluation gave its camera control a clear edge over two named rivals.
Reporting from 1 source: GIGAZINE.
World Labs has introduced Atlas, a world model that generates video, images, and 3D space from a single picture. It renders up to one minute of 1440p footage that follows camera movement and angle instructions. In human evaluations of camera control, Atlas beat Google's Gemini Omni Flash in 81 percent of comparisons and MiniMax's H3 in 75 percent. It also builds 3D space data through Gaussian splatting. Atlas is in limited release to partners, with early access expected within weeks.
Atlas accepts several reference images at once and combines them to reproduce a scene, and it handles plain text-to-image generation as well. The camera control is the headline feature: a user can specify movement and angle, and the model follows, which is how it produced the 81 percent win rate against Google's Gemini Omni Flash and 75 percent against MiniMax's H3 in human comparison.
For 3D output, Atlas takes one image, fills in the scenery the photo never showed, and forms space data using Gaussian splatting. World Labs is opening Atlas to select partners first; early access is planned within weeks, and interested users can join a waitlist.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.