News

World Labs Opens Atlas to Developers, Turning Still Images Into Scenes They Can Move Through

World Labs has released Atlas tools and developer APIs for generating and editing spatial scenes. Its documentation warns that unseen areas are invented.

World Labs is putting its world model into developers' hands, letting applications turn a collection of images into a scene users can move through and edit. Its November 12 release brings Atlas tools and developer APIs into workflows that include Blender.

The attraction is control over what happens beyond the original photograph. Instead of accepting a fixed camera angle, a creator can choose where to look next, generate that view and build a route through the scene. The model fills in what the source images do not show.

A scene that can leave the demo

The Atlas release includes Viewfinder for reconstructing scenes and rendering camera paths, Chisel for shaping their structure, and a tool that converts splats into meshes with geometry that other 3D software can use. Relight changes how splats are lit.

World Labs presents film sets, architectural spaces and robotics simulations as uses for these environments. Those are the company's proposed applications, rather than evidence that a generated scene reproduces a physical location accurately.

The missing room is still a guess

The developer documentation describes a sequence: position the starting images, attach camera and depth information, then request views from new camera positions. Returned views can become context for the next request. Masked edits let developers change selected areas while preserving the rest.

That loop also carries mistakes forward. World Labs tells developers to inspect generated views before adding them back, because errors in geometry or appearance can persist in later requests. It explicitly says Atlas invents areas outside the original images.

For creative work, that ability to supply the missing scenery is part of the appeal. For a reconstruction intended to describe an actual place, it is the boundary between what the camera captured and what the model made up.

Some offers on this page may be paid placements or contain affiliate links.

Subscribe to Techpresso

Free daily newsletter, read in 5 minutes.

Subscribe free