World Labs has opened access to its Atlas model to the public, launching two web applications and developer API endpoints. This release allows users to generate 3D scenes, images, and video clips from text prompts or reference images, moving the technology from a demonstration phase into a self-serve workflow.
The November 12, 2026, launch includes Viewfinder, which supports up to 64 input images for scene reconstruction, and Chisel, an agentic tool for 3D creation. While the company claims these tools support Gaussian splat exports, pricing, regional availability, and specific API request schemas remain unconfirmed.
World Labs announced on November 12, 2026, that it was opening Atlas access to everyone, releasing two web apps and the API endpoints behind them. The company says users can turn ideas or source images into keyframes, video clips, and 3D Gaussian splats.
The announcement’s central promise is straightforward: “We’re opening up access to Atlas to everyone today.”
Atlas itself is older than this release. World Labs introduced the model on September 1, 2026, describing a system that works across text, images, video, and 3D. November’s news is the broader access and tooling: Viewfinder for image-based world generation and reconstruction, Chisel for agentic 3D creation, and APIs for developers building their own interfaces.
That distinction matters. A model demonstration shows what a company says its technology can do. Apps and endpoints give creators and developers a way to start building workflows around it, subject to the access conditions and limits of those products.
Viewfinder Builds Worlds From One to 64 Images
World Labs describes Viewfinder as an image-to-world tool with two closely related uses:
Generate a world from one image. Atlas uses the image as a starting point and generates the surrounding scene.
Reconstruct an existing environment from up to 64 inputs. Additional views supply more information about the place being recreated.
Export images, videos, or 3D Gaussian splats. The company describes the generated images as spatially consistent.
A Gaussian splat is a form of 3D scene representation that can be rendered from different viewpoints. For a creator, the important distinction is the output’s intended use: an image gives you a frame, a video gives you a sequence, and a splat gives you a 3D representation to work with in a compatible workflow.
The single-image option is useful for expanding a visual concept. A reference picture can establish a room’s style, materials, and composition, while Atlas supplies views that the original image never showed.
Reconstruction has a different requirement: fidelity to something that already exists. World Labs’ September explanation explicitly says Atlas fills unseen areas with plausible content. More source images give it more evidence and reduce how much it needs to invent. The company illustrates this with a scene where a single garden photo leaves surrounding buildings unknown, while another image supplies information about a nearby cottage.
A plausible reconstruction is not the same thing as a verified record of a location. My reading is that Viewfinder’s one-image mode makes the most sense for creative expansion, while reconstruction work benefits from references that cover the areas the output needs to preserve. A convincing unseen wall is still generated content.
That is a practical distinction for anyone working from photographs of interiors, sets, or outdoor spaces. If a particular doorway or adjoining room matters, supplying a view of it gives the model evidence that a single attractive reference image cannot provide.
Chisel Starts With Prompts or Reference Images
Chisel approaches world creation through what World Labs calls agentic workflows for explicit 3D geometry. Users can begin with a text prompt or reference images, create a world in 3D, and then render images and videos or export splats. The company also says the workflow supports different visual styles in its November announcement thread.
The useful contrast is the starting point. Viewfinder is presented around images and reconstruction. Chisel is presented around directing the creation of a 3D world, with text as an available input.
For example, a creator with photographs of an existing environment has a clear reason to begin with Viewfinder. Someone developing an imagined setting from a written description has a more obvious starting point in Chisel. That is a workflow recommendation based on World Labs’ descriptions, not a hands-on comparison of output quality.
The phrase “explicit 3D geometry” also deserves care. It signals a workflow involving a 3D scene, but the announcement’s stated export options are images, videos, and splats. It does not describe editable mesh formats, topology controls, collision geometry, or direct game-engine integration.
I would check those requirements before treating Chisel as a source of production-ready game assets. The announced outputs already have uses in visualization and scene exploration; a downstream pipeline may need more than those outputs alone.
The Model’s Capabilities and the Apps’ Limits Differ
World Labs’ September model description and November app announcement operate at different levels. Reading them together reveals an important boundary: a capability described for Atlas is not automatically a documented setting in either web app.
Capability
September Atlas description
November access announcement
Reconstruction inputs
The model can use more than 100 input images in its spatial context
Viewfinder supports up to 64 inputs
Video output
Demonstrates up to one minute of video at 1440p
Lists video exports without specifying app duration or resolution limits
Camera control
Describes camera position and angle as model inputs
Does not detail the apps’ camera-control interfaces
Creation workflow
Covers generation, reconstruction, and simulation
Introduces Viewfinder, Chisel, and their underlying API endpoints
The 64-input figure is the clearest example. September’s post says Atlas can use more than 100 images; November’s Viewfinder description specifies up to 64. These statements are compatible if the app exposes a smaller input allowance than the underlying model supports. The published material does not explain the reason for the difference.
For planning purposes, I would use 64 as the documented Viewfinder ceiling, without assuming the API accepts the larger number. Model capacity, application limits, and endpoint limits are separate questions.
The same caution applies to video. The September post demonstrates a manually designed camera path producing a one-minute, 1440p video. That is a published model demonstration, not a promise that every newly available app mode exports that length and resolution.
These are vendor descriptions, not independent measurements. Their value here is in showing what World Labs has stated and where the newer product announcement leaves details unspecified.
The API Makes Atlas Available for Custom Workflows
World Labs says it is releasing all API endpoints powering Viewfinder and Chisel for developers and agents. The company’s announcement explicitly invites developers to build tools incorporating Atlas or let an agent interact with its models.
That opens a different route from working through a browser interface. A developer could build a project-specific front end, connect generation to an existing content pipeline, or have an agent coordinate requests as part of a larger task.
Potential workflows, based on the announced capabilities, include:
Reference-image intake: collect a scene’s source images and submit them through a custom tool.
Visual concept iteration: turn a prompt or reference set into images and clips for review.
3D export handling: route generated splats into a compatible visualization workflow.
Agent-directed creation: let software coordinate model calls around a user’s requested scene.
Those are integration possibilities, not documented endpoint recipes. The opened announcement does not supply request schemas, authentication steps, job-status behavior, or a list of supported parameters. There is no sound basis here for inventing a sample API call.
The consequential change is that developers no longer have to treat the web apps as the only interface World Labs is offering. The company says the models behind those apps are available programmatically. Whether a particular integration is practical will depend on the endpoint documentation, supported outputs, and operating conditions.
Open Access Leaves Cost and Regional Questions Unanswered
World Labs directs users to the Atlas sign-in page, but the cited announcement and public entry page do not provide a clear pricing schedule or usage quotas.
“Open to everyone” should therefore be read as the company’s access announcement, not as a promise of free or unlimited generation. The available evidence does not establish subscription tiers, included credits, per-generation charges, or API rate limits.
Those omissions matter differently depending on the job. A creator exploring a handful of scenes mainly needs to know the cost of iteration. A developer needs enough information to estimate repeated requests, unsuccessful jobs, and the cost of serving multiple users. An API’s existence alone cannot settle that budget.
Regional coverage is also unresolved. The opened sources do not provide a supported-country list or enough attributable evidence to verify specific reports of regional signup friction. It would be premature to promise uniform global signup availability.
Before committing a project, I would confirm three things through World Labs’ current product information: whether the account can register in the intended region, which output settings the chosen app or endpoint supports, and what usage will cost. Those checks determine whether Atlas is merely available to explore or suitable for a repeatable production workflow.
Frequently Asked Questions
3 questions
1
Where Can I Start Using Atlas?
World Labs directs users to the Atlas entry page. Its announcement names Viewfinder and Chisel as the web apps and says developers can use the endpoints powering them.
2
Does Atlas Require Special Capture Equipment?
World Labs’ September model description says Atlas can reconstruct scenes without special capture equipment or hundreds of densely captured views. The November announcement specifies that Viewfinder accepts up to 64 input images.
3
Did World Labs Announce a New Atlas Version?
The November announcement is an access-and-tooling expansion. Atlas was introduced on September 1, 2026; the later announcement names new apps and available endpoints without identifying a new numbered model version.
