World Labs Opens Its Atlas World Model to Everyone, but Still Lists No Price
Fei-Fei Li's startup released two web apps and a public API that turn photos or rough layouts into explorable 3D scenes. Its quality claims remain company-reported, and there is no model card or rate card yet.
Published
Last updated / Fact-checked on
World Labs opened Atlas, its next-generation world model, to the public on November 12, ending a partner-only period that began in September, AI Weekly reported. Anyone can now sign in and turn photos or a rough block-out into a 3D scene they can move a camera through, then export video, keyframes or 3D Gaussian splats.
Photo: Pixabay. Image is illustrative.
The company's blog lists the November 12 post as "Atlas: Releasing our next generation world model," saying users can "generate and reconstruct worlds and bring creations into your own workflows with our tools and developer APIs."
The release at a glance
| Item | Detail |
|---|---|
| Company | World Labs (led by Fei-Fei Li) |
| Model | Atlas, first announced Sept. 1, 2026 |
| Public release | Nov. 12, 2026 |
| Apps | Viewfinder (photos to 3D), Chisel (layout to 3D) |
| Extra tools | Splats to Mesh, Relight |
| Developer access | Public API, Blender integration |
| Real-time mode | Frameboy research preview, up to 15 fps |
| Price | Not published |
Sources: World Labs; Genativity; Tested Intelligence; Screenweaver.
What Atlas is
World Labs introduced Atlas on September 1 as an "omni model" pretrained from scratch on text, images, video and 3D. The company calls it "a multimodal autoregressive diffusion transformer," meaning all inputs feed a shared spatial context that the model uses to generate what comes next while staying consistent in 3D.
The company says Atlas can generate up to one minute of 1440p video from reference images with "pixel-perfect camera control," reconstruct real scenes from one to dozens of photos, re-frame existing videos and produce images and 360-degree panoramas from text. It also says Atlas supports "Real-to-Sim workflows for robotics," where a real environment is turned into a simulation a robot can practice in.
What shipped on November 12
Two browser apps run on the same model, according to Genativity and Screenweaver:
- Viewfinder rebuilds a real place from as many as 64 overlapping photos, or imagines the surroundings from a single image. Users can explore new viewpoints and render camera paths.
- Chisel works from geometry first. Users block out walls and objects from a text prompt or reference images, move things around, and every camera view updates before rendering.
Two more tools came with the release: Splats to Mesh, which converts Gaussian splats into meshes with real geometry for game engines, and Relight, for changing lighting on splats, according to Tested Intelligence.
The developer API no longer requires a partner agreement, Genativity reported, and includes endpoints for scene generation, video, camera posing and 3D reconstruction. World Labs pitches it at AI agents and existing pipelines, including Blender.
A real-time variant called Frameboy, released November 5 as a research preview, streams generated worlds from cloud GPUs to a browser at up to 15 frames per second while the user steers the camera, according to FourWeekMBA. World Labs itself said Frameboy "is not cheap compared to local rendering methods."
What is still missing
Three things did not ship with the public release.
Pricing. None of the Atlas pages list a cost per generation, according to Genativity, Tested Intelligence and Screenweaver. For a studio or developer, that is the number needed to plan a budget.
A model card or paper. World Labs has not published an academic paper or model card for Atlas, Genativity reported.
Independent checks. In September, World Labs said human raters preferred Atlas 75% to 94% of the time over five rival video models in blind tests of camera-controlled generation. Those figures are the company's own, Genativity noted, and no outside group has reproduced them. Secondary accounts also disagree slightly on the range.
Why it matters
World models are a step beyond image and video generators. Instead of producing a flat clip, they build a scene that stays consistent as the camera moves. That makes them useful for film pre-visualization, game level design, architecture walk-throughs and robot training.
Until recently, most world models were lab demos or closed partner programs. Atlas now has a front door anyone can use, plus an API and export formats that fit existing 3D tools. That puts it in front of creators and developers who can test the claims themselves.
The release also matters for robotics. Training robots in simulation is cheaper and safer than in the real world, but building accurate simulations is slow. A model that turns a few photos of a warehouse or kitchen into a usable 3D scene could cut that work, if the geometry is accurate enough.
Corporate context adds weight. AMD agreed in September to acquire World Labs in an all-stock deal valued at about $8.2 billion, with work starting on training and inference on AMD GPUs, FourWeekMBA reported, citing World Labs. A public product with a growing user base strengthens AMD's position in AI software, not only chips.
What to watch
- A rate card. Whether World Labs publishes per-generation or subscription pricing.
- Independent tests. Whether researchers or studios publish comparisons against rival world models.
- Robotics use. Whether robot makers adopt Atlas for real-to-sim training.
- AMD integration. How the acquisition changes where and how Atlas runs.
More coverage is on our AI News hub.
How we reported this
We reviewed World Labs' blog and its September 1 Atlas post, and compared four independent reports on the November 12 release. Performance figures are the company's and are labeled as such. We did not test Atlas ourselves. The image is a stock photo, not a World Labs product image.
Published and fact-checked November 13, 2026.
Sources
- World Labs, "Atlas: A World Model for Spatial Intelligence," Sept. 1, 2026: https://worldlabs.ai/blog/atlas
- World Labs, Research & Insights (Nov. 12 release listing): https://worldlabs.ai/blog
- Genativity, "World Labs opens Atlas to everyone with two web apps and a public API, and still lists no price," Nov. 13, 2026: https://genativity.com/news/world-labs-atlas-public-access-no-published-price
- Tested Intelligence, "World Labs Opens Atlas to Everyone: Photos In, Explorable 3D Worlds Out," Nov. 12, 2026: https://testedintelligence.com/world-labs-atlas-open-to-everyone/
References
- AI Weekly, "World Labs Opens Atlas World Model to Public With Viewfinder, Chisel and Developer APIs," Nov. 13, 2026: https://aiweekly.co/alerts/world-labs-opens-atlas-world-model-to-public-with-viewfinder-chisel-and
- Screenweaver, "World Labs Opens Atlas: Build a 3D Set From Location Photos," Nov. 13, 2026: https://www.screenweaver.ai/blog/world-labs-atlas-3d-sets-previz
- FourWeekMBA, "World Labs, Joining AMD in $8.2B Deal, Streams Worlds Live," Nov. 6, 2026: https://fourweekmba.com/ai-world-labs-joining-amd-in-8-2b-deal-streams-worlds-live/
- Image: Pixabay, Pixabay Content License: https://pixabay.com/photos/virtual-reality-4490469/
DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.