World Labs Unveils Atlas Multimodal World Model

World Labs, founded by AI scientist Fei-Fei Li, is building systems designed to understand and simulate three-dimensional environments rather than merely generate text or flat images. World models are increasingly viewed as foundational technology for robotics, embodied AI and interactive content because machines must grasp spatial relationships, physical change and the consequences of actions to operate reliably in real or virtual settings.

World Labs has unveiled Atlas, a new multimodal world model that natively handles four data types: text, images, video and 3D content. The model supports camera-controlled generation, spatial reconstruction and simulations across space and time, including representations of real-world environments and robotic operations. The release positions Atlas as a broader platform for spatial intelligence, combining generation, reconstruction and simulation capabilities within a single general-purpose model.