World Labs unveils Atlas, an omni world model for spatial intelligence with pixel-perfect generation

World Labs just pulled the curtain back on Atlas, a model the company is calling an “omni world model” for spatial intelligence. It generates and reconstructs 3D images and videos with full camera control, pushing the boundaries of what AI can do when it actually understands physical space. The model, unveiled on September 1, 2026, represents a major leap for the startup co-founded by Stanford AI luminary Fei-Fei Li. Atlas can produce pixel-perfect, camera-controlled imagery and video at resolutions up to 1440p, running for durations of up to one minute. What Atlas actually does Technically, Atlas is a multimodal autoregressive diffusion transformer. In plainer terms, it’s a single model that can process and generate across text, images, video, and 3D data simultaneously, all within a unified spatial framework. World Labs claims the model’s spatial reconstruction capabilities outpace existing specialized 3D models. The road from Marble to Atlas Atlas didn’t materialize out of thin air. It builds on Marble, World Labs’ first commercial multimodal model, which went public in November 2025. Marble supported 3D world creation from multimodal inputs and served as the proving ground for ...