Table of Contents
World API by World Labs
The World API is a developer platform for “Marble,” a spatial intelligence model that generates explorable, persistent 3D worlds from 2D inputs. Released in January 2026 by World Labs (founded by Fei-Fei Li), it brings “Large World Model” capabilities to third-party applications.
What It Is
The World API provides programmatic access to Marble, a generative AI model designed to understand and create 3D space. Unlike 3D asset generators that create isolated objects, Marble builds complete, coherent environments. Developers send text descriptions, reference images, or video clips to the API, and the model constructs a 3D representation (often represented as Gaussian Splats or textured meshes) that users can navigate through, offering consistent geometry and lighting rather than just a static 360° image.
Key Features
- Marble World Model: Powered by World Labs’ proprietary “Large World Model” (LWM), which understands 3D physics and spatial relationships better than standard diffusion models.
- Multimodal Inputs: Generates worlds from Text (prompts), Images (single or multi-view), and Video (translating camera movement into 3D depth).
- Explorable Outputs: Unlike 2D skyboxes, generated worlds have depth and geometry, allowing for actual movement and parallax within the scene.
- Export Formats: Supports exporting environments as Gaussian Splats (.ply) for high-fidelity rendering or standard Meshes (.obj/.glb) for game engine integration.
- Quality Tiers: Offers varying compute tiers (e.g., “Draft” for rapid iteration, “Standard” for final production) to manage latency and cost.
Use Cases
- Game Level Generation: Rapidly blocking out explorable 3D maps and environments for gaming prototypes.
- Virtual Production: Creating photorealistic 3D backdrops for film/video that maintain consistency across different camera angles.
- Architectural Visualization: Transforming sketches or reference photos into walkthrough-ready spatial pre-visualizations.
- Immersive Commerce: Generating 3D branded spaces or storefronts from product imagery.
Pros and Cons
- Pros: Delivers true “Spatial Intelligence” with consistent geometry, moving beyond the “dream-like” inconsistency of video generators; flexible input options allow for precise art direction; backed by top-tier research (Fei-Fei Li).
- Cons: High cost per generation (approx. $1.20 for Standard quality) makes it expensive for casual use; generation is not real-time (latency involves processing time); “Draft” quality may lack the texture fidelity required for final production.
Pricing
- Credit-Based System: Operates on a pay-as-you-go model.
- Rate: Approximately $1 USD per 1,250 credits.
- Cost Per Gen: A “Standard” quality world generation typically costs ~1,500 credits (approx. $1.20), making it a premium service compared to basic asset generators.
How Does It Compare?
- Luma AI (Genie/Dream Machine): The closest competitor. Luma excels at object generation and video-to-3D. Comparison: Luma is generally faster and better for individual assets or short video clips, whereas World Labs focuses on coherent, navigable environments (scenes).
- Blockade Labs (Skybox AI): Generates 360° panoramic images. Comparison: Skybox creates a 2D sphere (you can look around but not walk into the scene). World API creates volumetric 3D data (you can move through the space), offering true 6-DOF (Six Degrees of Freedom) movement.
- CSM (Common Sense Machines) / Meshy: Specialized in 3D object/character generation. Comparison: Use these tools to generate a “chair” or “character”; use World API to generate the “room” or “forest” they inhabit.
- Odyssey: A high-end visual AI competitor. Comparison: Odyssey focuses on “Hollywood-grade” visual fidelity for video and scenery, often competing on texture quality, while World Labs emphasizes the structural/spatial logic of the world.
Final Thoughts
The World API represents a shift from “Generative Media” (pixels) to “Generative Reality” (spatial data). By solving the consistency problem—ensuring a chair looks like the same chair from behind—Marble bridges the gap between AI video and game engines. While the pricing (~$1.20/gen) places it firmly in the B2B/Prosumer category, it offers a unique capability: the ability to “step inside” an image. For developers building the next generation of spatial apps (XR/VR), this API provides the foundational layer of world creation that was previously manual and labor-intensive.