World Labs Unveils Atlas World Model
Atlas generates camera-controlled video and reconstructs 3D scenes from input images.
World Labs announced Atlas, its multimodal world model for spatial intelligence. The company states Atlas generates image and video frames with pixel-perfect camera control and reconstructs explicit 3D output from one or more input images. Researcher Gowthami Somepalli and engineer Ben Mildenhall described it as an autoregressive diffusion model for next-frame prediction that also supports novel view synthesis. World Labs said passing more images reduces reliance on generated content. The firm plans early access for select partners and posted a technical blog at worldlabs.ai/blog/atlas.
Combined views
5.3M
43 posts, first seen 13h ago
