World Labs

Atlas is World Labs' new omni world model, aimed at teams building in graphics, robotics, and simulation who need a single system that can generate images and video, reconstruct 3D scenes, and let a user move a camera through space with precise control. World Labs describes Atlas as "a multimodal autoregressive diffusion transformer," pretrained from scratch, with camera geometry treated as a native input rather than something coaxed out of a text prompt . In practice, it produces images and videos with precise camera control, including videos up to one minute at 1440p , and reconstructs scenes as 3D Gaussian splats from as few as one image up to dozens. The launch matters because it consolidates work that previously required separate video, reconstruction, and simulation pipelines. The new world model is designed to replace handoffs among video, reconstruction and simulation tools, using images, camera poses, and depth maps to generate controlled video, reconstruct scenes, and produce simulated robot-camera observations . Atlas builds directly on the company's prior work, including Marble, World Labs' first commercial multimodal model, which went public in November 2025 , and follows the July 2026 acquisition of SceniX to deepen robotics coverage. The company was founded in January 2024 by Fei-Fei Li (Co-Founder, CEO), Justin Johnson, Christoph Lassner, and Ben Mildenhall to develop 3D environment generation, complete with real-time simulation , a group with deep roots in computer vision, generative modeling, and neural rendering. Atlas is entering early access with select partners , so the immediate audience is founders and research teams who want to prototype spatially grounded applications, from film and game previsualization to robot policy training, against a single model rather than stitched-together components.