World Labs just made camera control native to world models
Signals InboxSeptember 2, 2026World Models
World Labs launched Atlas, a world model that takes camera geometry as a native input and can generate up to one minute of 1440p video while reconstructing scenes in 3D. Video generators have been good at making plausible motion. Atlas is trying to make the virtual camera obey exact spatial instructions.
Introducing Atlas:
— World Labs (@theworldlabs) September 1, 2026
The world's first multimodal world model that generates image and video frames with pixel-perfect camera control and reconstructs them in 3D.
Model the world, move the camera, and simulate space & time. pic.twitter.com/o0qeGubi19
Q1What did World Labs officially launch?
World Labs' official Atlas announcement describes a multimodal autoregressive diffusion transformer trained to operate natively across text, images, video and 3D.
Q2What does pixel-perfect camera control mean?
Atlas accepts precise camera geometry rather than only text such as pan left or zoom in. A user can specify a camera path and generate views that follow that trajectory while maintaining scene geometry.
Q3How much video can it generate?
World Labs says Atlas can output up to one minute of video at 1440p from one to six reference images, depending on the task and camera path.
Q4Why does 3D reconstruction matter?
A model that reconstructs explicit spatial structure can be useful beyond entertainment, including robotics simulation, virtual environments and real-to-sim workflows where machines train on recreated physical spaces.
Q5What is the market tension?
Creative video tools optimize for visual quality, while robotics needs spatial consistency and controllability. Atlas is trying to serve both, which could make world models a shared infrastructure layer for media and physical AI.