The Pulse
World Labs opens Atlas world model through a public API
Developers can use Atlas to generate and edit spatial scenes through public API endpoints. World Labs also offers tools for reconstructing scenes, shaping worlds and converting splats into meshes.

Developers can now use World Labs’ Atlas world model through public API endpoints to build and edit spatial scenes in their own applications. World Labs lists the release of Atlas tools and developer APIs on November 12, alongside integrations with existing tools including Blender.
Atlas turns images into explorable, editable 3D worlds. The release also includes tools for reconstructing scenes, shaping a world and converting splats into meshes.
Tools for building scenes
Viewfinder reconstructs a scene from photographs or generated images, letting users explore different viewpoints and render camera paths. Chisel lets users shape a world through controllable scene building and grayboxing.
The company also lists Splats to Mesh for converting splats into meshes for games and interactive environments, and Relight for manipulating the lighting of splats.
Inside the API workflow
The API workflow starts by positioning images with a task called images2PosedRGBD, according to World Labs’ documentation. Images with overlapping views can be submitted together so the service can estimate their relative positions, while separately posed images can be placed in a shared scene for a creative composition.
The resulting views include a camera and a depth map for each image. Developers can then send those views to atlasGenerate with target cameras specifying where to create new views.
Atlas returns an RGB image and camera for each target, with depth available on request. Those returned views can be used as context for further requests, while an ordered camera path can be submitted to make a video or flythrough.
Another task, atlasChisel, makes masked edits across posed views. The documentation describes using separate masks to mark the parts of images to edit while preserving the rest.
How Atlas models a world
World Labs introduced Atlas on September 1 as a world model for spatial intelligence. The model was pretrained to operate on text, images, video and 3D, and combines its inputs into a shared spatial context.
For spatial reconstruction, Atlas can use multiple input views and may invent parts of a scene that are not visible in them. In its introduction to the model, World Labs said it could generate images and videos from one or more images with camera control, outputting up to 1 minute of video at 1440p.
The model can also generate new views and estimate their geometry, producing outputs such as point clouds and 3D Gaussian splats. For video of a real space, the company says Atlas can predict the depth of each frame and combine the results into a 3D reconstruction.
Sources
Companies in this story
Related stories
- AMD to Buy Fei-Fei Li’s World Labs for $8.2 Billion
- AMD to Acquire World Labs in $8.2B All-Stock Deal
- Salesforce releases open-weight speech transcription model
- Anthropic advises deleting "think hard" lines from Claude Opus 5.5 prompts
- AWS makes snapshot-based AgentCore V2 the default for new runtimes in select regions