Atlas: World Labs' Omni World Model for Spatial Intelligence

World Labs has introduced Atlas, an "omni world model" for spatial intelligence. Atlas is pretrained from scratch to natively operate on text, images, video, and 3D, combining all inputs into a shared spatial context. It's a multimodal autoregressive diffusion transformer that generates outputs consistent with the 3D geometry it's seen, and can even imagine what lies beyond the input frames.

What Atlas Can Do

Atlas is designed to handle three core tasks: world generation, reconstruction, and simulation.

Camera-Controlled Generation

Atlas generates images and videos from one to six reference images with precise camera control. You specify the camera path—no text-based prompts needed. It outputs up to 1 minute of video at 1440p. From a single image, it can extrapolate unseen parts of a scene (e.g., the back side of a robot, a lawn next to a pool).

Spatial Reconstruction

Atlas reconstructs real-world scenes from one to dozens of input images. According to World Labs, it outperforms state-of-the-art 3D reconstruction models and produces both novel view frames and explicit 3D outputs. This is a step forward on the decades-old problem of novel view synthesis from sparse images.

Space-Time Simulation

Atlas models space and time from input videos. It can re-frame videos for visual effects and supports Real-to-Sim workflows for robotics.

Image Generation

From text, Atlas can generate images and 360 panoramas, following complex prompts and rendering text with a variety of visual styles.

Key Technical Details

The model works by encoding inputs into a "spatial context"—each image is grounded at a 3D position. This context allows you to place unrelated images at arbitrary 3D locations and Atlas will generate a world that interpolates between them (imagining doorways, hallways, etc.). This enables long, controllable videos: "you are staging the scene, not pulling the lever of a slot machine."

Who It's For

Developers working in 3D content creation, robotics simulation, or anyone needing controllable video generation from sparse image inputs. Atlas will power future versions of World Labs' Marble product. Early access is available on the World Labs site.