World Labs opens Atlas world model to everyone
Viewfinder rebuilds a real place from up to 64 photos, generates a whole world from one image, and exports video or 3D Gaussian splats.

Key takeaways
- World Labs opened Atlas to the public on November 12, 2026, with the Viewfinder and Chisel web apps and the API endpoints behind them.
- Viewfinder reconstructs a real place from up to 64 photos or generates a whole world from one image, and exports images, video clips or 3D Gaussian splats.
- Atlas outputs leave as Gaussian splats in SPZ or PLY format, or convert to meshes for Blender and game engines.
- World Labs has not published Atlas pricing, and the documentation and API hostnames still carry beta labels.
World Labs opened its Atlas world model to everyone on November 12, 2026, moving the system out of partner-only early access and shipping two web apps, a set of supporting tools and a public developer API.
The company announced the release on its Atlas page and on X, saying users can "bring your ideas for building a 3D world and come away with keyframes, video clips, or 3D Gaussian splats." The endpoints behind both apps are public, documented at docs.atlas.worldlabs.ai.
Viewfinder works from images. A single photo generates a plausible surrounding scene, and as many as 64 overlapping photos reconstruct a real environment. Chisel works from geometry: you assemble planes and blocks into walls, furniture and structure, then style the scene with text prompts or reference images while an agent applies a sequence of revisions to the layout. Both run on the same Atlas model and export keyframes, video clips and 3D Gaussian splats.
Atlas is World Labs' omni world model, introduced on September 1, 2026. The company pretrained it from scratch to work natively on text, images, video and 3D, and describes it as a multimodal autoregressive diffusion transformer.
In its September model post, World Labs says Atlas generates up to one minute of video at 1440p with pixel-perfect camera control from a small number of reference images, and reconstructs scenes from one to dozens of images. The company also says Atlas outperforms specialized open-source models on 3D reconstruction benchmarks, a result that AlphaSignal notes has not been independently replicated.
Reconstructed worlds can leave as 3D Gaussian splats, exportable in SPZ or PLY format, or convert into meshes for Blender and game engines. AlphaSignal lists the public API operations as atlasGenerate, atlasTextToImage, atlasChisel, atlasMasked, images2PosedRGBD and splats2Mesh. Atlas also models space and time from input video, which World Labs says supports visual-effects reframing and robot simulation.
Pricing is the open question. World Labs has no dedicated Atlas pricing page, and the documentation and API hostnames still carry beta labels, as VP Land and AlphaSignal report. AlphaSignal points to Marble Studio's separate credit schedule as a rough reference, but says Atlas workloads may be priced differently.
For creators, the change means a set of location photos or a rough gray-box layout can become a navigable 3D scene and leave as a splat or a mesh without a partner agreement. The open questions are what Atlas costs and how faithfully it rebuilds the parts of a scene no camera saw.
Atlas is available now at atlas.worldlabs.ai; the API walkthrough starts at docs.atlas.worldlabs.ai/quickstart.
Sources
- worldlabs.ai - November 12 release page; the tools and the API
- worldlabs.ai - September 1 model post; specs and benchmark claims
- x.com - company announcement, November 12
- docs.atlas.worldlabs.ai - Atlas API documentation
- vp-land.com - launch coverage; pricing and beta-label report
- alphasignal.ai - API operations, pricing gap, benchmark caveat
- tokenpost.com - launch coverage
- radiancefields.com - splat and mesh tasks, Blender