World Labs releases Atlas world model with 3D reconstruction and camera trajectory control
- World Labs released Atlas, which it describes as a multimodal world model that generates image and video frames, reconstructs scenes in 3D, and accepts camera control.
- Atlas takes camera trajectories and geometry as native inputs, which World Labs says allows precise control of perspective rather than relying only on text prompts for camera movement.
- World Labs says Atlas can create detailed simulated 3D environments from a single image and render them from different camera angles.
- The model uses a multimodal autoregressive diffusion transformer architecture, according to World Labs, and is part of its effort to build physics-aware world models for spatial reasoning.
- Fei-Fei Li co-founded World Labs in February 2024; the company has raised $1.2 billion from backers including Nvidia, AMD, and Autodesk.