Atlas by World Labs is an omni world model designed for spatial intelligence, capable of generating, reconstructing, and simulating 3D environments from multimodal inputs like text, images, and video.
What the product does and how it is positioned
Atlas is a multimodal autoregressive diffusion transformer developed by World Labs. It is engineered to understand the appearance, behavior, and evolution of worlds, enabling users to generate consistent 3D content and simulate real-world environments.
The model integrates spatial context by grounding inputs in 3D space, allowing for precise camera control and the synthesis of novel views. It is designed to scale with increased training compute, aiming to provide high-fidelity outputs for creative, robotics, and simulation workflows.
Source-supported ways to use the product
Enables Real-to-Sim workflows by reconstructing physical spaces and generating sensor data, such as RGB and depth, to train and test robot navigation and manipulation.
Allows creators to reframe videos from multiple angles, freeze time, and generate consistent long-form video content using precise camera paths.
Atlas departs from standard LLM and video model architectures by utilizing a multimodal autoregressive diffusion transformer. This design treats spatial control as a core component by conditioning all inputs and outputs on explicit camera poses and 3D positions.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
Atlas is a multimodal model that natively operates on text, images, video, 3D depth maps, and camera poses to form a shared spatial context.
No, Atlas can reconstruct real-world spaces from as few as one to dozens of ordinary images, eliminating the need for dense capture equipment.
Atlas uses precise camera geometry as a native input type, allowing users to specify exact camera positions and angles rather than relying on text-based instructions.
Yes, the model can output explicit 3D representations, including point clouds and 3D Gaussian splats, which are suitable for robotics, gaming, and design workflows.
Yes, Atlas supports Real-to-Sim workflows by reconstructing environments and generating the RGB and depth data that a robot's sensors would observe during navigation or manipulation.
