Overview
World Labs Atlas is an omni world model for generating, reconstructing, and simulating spatially consistent environments. World Labs introduced it on September 1, 2026 as a model pretrained from scratch to work natively with text, images, video, camera poses, depth maps, and 3D data.
Atlas is more than a conventional AI 3D model generator. It can create controlled camera paths through generated or reconstructed scenes, infer unseen parts of an environment, produce explicit 3D representations, and model how a scene changes over time. These capabilities target creative production, VFX, game development, design, robotics research, and simulation.
Atlas is a separate model from World Labs' Marble product. The company says Atlas will power future versions of Marble and other World Labs products, but Atlas itself currently has its own limited-access program rather than a generally available consumer interface.
Key Features
Camera-controlled generation - Atlas accepts one or more reference images plus explicit camera positions and angles, then generates new views that follow the requested path. World Labs demonstrates videos up to one minute at 1440p.
Sparse-view spatial reconstruction - The model reconstructs scenes from one to more than 100 input images. With limited input it fills unseen regions plausibly; additional views reduce how much geometry it needs to infer.
Explicit 3D output - Atlas can produce depth-aware novel views, point clouds, and 3D Gaussian splats for workflows that need a scene representation rather than only rendered frames. Teams comparing reconstruction approaches can also review Meta SAM 3D.
Space-time simulation - Atlas models spatial structure and temporal change, enabling video reframing, multiview "bullet time" effects, and Real-to-Sim environments for robot navigation and manipulation.
Multimodal spatial context - Text, images, camera poses, and 3D depth maps are combined into a shared spatial context. This lets the model condition later outputs on both appearance and camera geometry.
Image and panorama generation - Atlas generates images and 360-degree panoramas from text or image prompts, including varied visual styles and text rendering, although World Labs describes world modeling as its primary focus.
Pricing & Plans
World Labs has not announced public pricing, usage allowances, or a self-service Atlas plan. The model is entering early access with select partners, and prospective users must submit the official access request form.
| Access item | Current status | What to verify |
|---|---|---|
| Atlas model access | Select-partner early access | Eligibility, onboarding timeline, and supported use case |
| Public pricing | Not announced | Commercial terms, usage limits, and compute charges |
| Self-service product | Not announced | Interface, API availability, exports, and collaboration features |
| Marble integration | Planned for future versions | Timing and which Atlas capabilities will be exposed |
Do not treat the absence of published pricing as a free tier. Teams should wait for written access terms before estimating production costs or making architecture commitments.
Best For
- VFX and virtual production teams that need precise camera paths through generated or reconstructed scenes.
- 3D teams comparing world-level reconstruction with asset-focused products such as Luma AI Genie.
- Robotics groups building Real-to-Sim environments for navigation or manipulation research.
- Game and simulation developers exploring spatially consistent scene generation.
- Research partners able to evaluate a gated model before public pricing and API guarantees are available.
Creators who only need finished clips should compare the broader AI video generator category and the current AI video generator guide. For standalone visual generation, the AI image generator guide covers products that are already available to a wider audience.
FAQ
What is World Labs Atlas?
Atlas is World Labs' next-generation omni world model for generating, reconstructing, and simulating spatial environments. It works across text, images, video sequences, camera poses, depth maps, and explicit 3D outputs.
Is Atlas a new version of Marble?
No. Atlas is a separate foundation model. World Labs says it will power future versions of Marble and other products, but the company has not presented Atlas as a numbered Marble release.
Can anyone use Atlas now?
No. As of September 3, 2026, Atlas is in early access with select partners. Interested teams can request access, but World Labs does not promise acceptance or a public onboarding timeline.
How much does Atlas cost?
World Labs has not published Atlas pricing. There is no verified free plan, paid tier, API rate, or usage allowance to report yet.
What can Atlas generate?
Atlas can generate camera-controlled images and videos, novel views of scenes, 360-degree panoramas, depth-aware outputs, point clouds, and 3D Gaussian splats. World Labs also demonstrates video reframing and robotics simulation workflows.
How long can Atlas videos be?
World Labs says Atlas can output camera-controlled video up to one minute at 1440p. Actual limits for early-access partners may depend on the access agreement and implementation.
Does Atlas faithfully reconstruct unseen areas?
Not necessarily. When input views do not cover part of a scene, Atlas uses learned world knowledge to generate a plausible continuation. Providing more input images gives it more evidence and reduces the amount it must invent.
Does Atlas have an API?
World Labs operates a public World API for Marble-related world generation, but the Atlas announcement does not document a generally available Atlas API, model identifier, or public rate limits. Teams should not assume that current World API access includes Atlas.
Are Atlas benchmark claims independently verified?
The launch article reports World Labs' evaluations for camera-conditioned generation and 3D reconstruction, including third-party human raters for one comparison. These remain vendor-published results until broader independent evaluations are available.