Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)

World Labs Opens Atlas World Model to Everyone and Releases 3D Scene API

World Labs has removed the waitlist: the Atlas world model is now available to everyone, and developers get an API. In Viewfinder, you upload photos, Atlas fills in the space beyond the frame, and you can place the camera at any point. The result is saved as Gaussian Splatting and converted into meshes for Blender and game engines. For studios and product teams, this means a quick 3D scene draft from a couple of shots.

AI-processed from TG: @ai_machinelearning_big_data (Machinelearning); edited by Hamidun News
World Labs Opens Atlas World Model to Everyone and Releases 3D Scene API
Source: TG: @ai_machinelearning_big_data (Machinelearning). Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTG: @ai_machinelearning_big_data (Machinelearning) ↗
Editor-in-Chief
Chisel works in reverseit generates images from a simple 3D blockout
Key metric #2Previously, users had to request custom scenes, and finished ones could only be viewed
Key metric #3The startup itself is set to join AMD, with the deal expected to close by the end of the year

Generating 3D worlds has long been a toy for those who managed to get an invitation. Now World Labs has lifted that restriction: the Atlas world model is open to anyone who wants it, and developers get an API. Through it, scene generation, video, and 3D reconstruction can be built into your own product. For those making games, architectural visualizations, ads, or assistants for designers, this is a reason to test the tool on a real task.

What a world model is and how to work with it

A world model is a neural network that builds not a single image but an entire space. From an ordinary photo, it fills in what isn't in the frame: what sits around the corner, what the room behind the photographer looks like. Then you can move through that space and view it from different points.

World Labs offers two web tools. The first is called Viewfinder. You upload one or several photos, Atlas extends the scene beyond the frame, and then you can place the camera at any point. From there, you capture images and video from different angles. Imagine you have a couple of interior shots for an upcoming video. Instead of modeling by hand, you get a space in which the director chooses the angles.

The second tool, Chisel, works in the opposite order. First, you assemble a simple 3D scene: you place walls and objects, like in a construction kit. Then the model generates images based on this blockout. This way you control the layout and composition, while the neural network handles the details and style. For a designer or level artist, this is closer to familiar work than a prompt of a single line.

What the output looks like

The result format matters most in practice. The scene is saved in Gaussian Splatting, meaning a cloud of small semi-transparent "splats" that render quickly and look photorealistic. But this format is awkward to work with in standard pipelines. So the scene can be converted into polygonal meshes with real geometry. Such meshes open in Blender and game engines. That means Atlas output doesn't remain a picture in the browser: you can refine it, set lighting, add objects, and build it into a game or interactive demo.

For developers, the API is a separate piece of news. It lets you avoid sending users to a third-party site and instead run scene generation and reconstruction inside your own service. For example, an app for real estate agents could build a walkable space from several photos of an apartment. This is an illustrative possible scenario, not a case the company has announced. Access terms and pricing were not disclosed in the original announcement, so check them on the Atlas website before planning a product.

Context and what's next

Atlas was introduced in September. In early November, the company announced Frameboy, a tool for exploring virtual worlds in real time. Until now, in the public version, finished scenes could only be viewed, and creating your own in Viewfinder was allowed only through a testing request. Now Viewfinder is open to everyone, and a world built from your own photos can be walked through immediately in Frameboy. The combination is complete: upload a photo, get a space, and walk through it.

There is also corporate news. The startup itself is set to become part of AMD, with the deal expected to close by the end of the year. For developers, this means uncertainty: it is not yet clear how the API terms, pricing, and development plans will change after the deal closes. If you plan to build a product on Atlas, it is worth preparing a fallback in advance, or at least not tying your entire architecture to a single service.

Practical takeaway

The opening of access is a good moment for a quick test. Take 3–5 photos of your object or a sketch of a scene, run them through Viewfinder and Chisel, and check the key points: how cleanly the meshes open in Blender or your engine, and how closely the generated space matches reality. The original announcement contains no quality comparisons or figures, so you will have to evaluate the results yourself. For prototypes, previsualization, and interactive demos, the tool looks useful. For final production, first make sure the geometry and details meet your standards.

Story

Key Questions Answered

What a world model is and how to work with it?

A world model is a neural network that builds not a single image but an entire space. From an ordinary photo, it fills in what isn't in the frame: what sits around the corner, what the room behind the photographer looks like. Then you can move through that space and view it from different points. World Labs offers two web tools. The first is called Viewfinder. You upload one or several photos, Atlas extends the scene beyond the frame, and then you can place the camera at any point.

What the output looks like?

The result format matters most in practice. The scene is saved in Gaussian Splatting, meaning a cloud of small semi-transparent "splats" that render quickly and look photorealistic. But this format is awkward to work with in standard pipelines. So the scene can be converted into polygonal meshes with real geometry. Such meshes open in Blender and game engines. That means Atlas output doesn't remain a picture in the browser: you can refine it, set lighting, add objects, and build it into a game or interactive demo.

What are the key details on Context and what's next?

Atlas was introduced in September. In early November, the company announced Frameboy, a tool for exploring virtual worlds in real time. Until now, in the public version, finished scenes could only be viewed, and creating your own in Viewfinder was allowed only through a testing request. Now Viewfinder is open to everyone, and a world built from your own photos can be walked through immediately in Frameboy. The combination is complete: upload a photo, get a space, and walk through it. There is also corporate news.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…