2025 · World Labs · Intermediate
Marble: A Multimodal World Model
Marble generates persistent 3D worlds from text, images, video, panoramas, or coarse 3D layouts.
Direct answer
What does Marble: A Multimodal World Model contribute?
Marble generates persistent 3D worlds from text, images, video, panoramas, or coarse 3D layouts.
Background
World Labs presents Marble as a multimodal world model for reconstructing, generating, editing, and exporting spatially consistent 3D worlds. It is product-facing, but directly represents the spatial-intelligence research direction.
Problem
The work addresses a central constraint in Spatial Intelligence: building systems that learn useful representations or actions while remaining general enough to transfer beyond a single demonstration or environment.
Core idea
Marble generates persistent 3D worlds from text, images, video, panoramas, or coarse 3D layouts.
Architecture and method
World Labs presents Marble as a multimodal world model for reconstructing, generating, editing, and exporting spatially consistent 3D worlds. It is product-facing, but directly represents the spatial-intelligence research direction.
- Multimodal 3D world generation
- Interactive world editing
- Exportable Gaussian splats, meshes, and videos
Results and impact
It makes the world-model layer visible to builders: generated 3D environments can become creative tools, simulation assets, and eventually training contexts.
Prerequisites
- 3D representations
- Generative models
- World models
Recommended reading order
Read the explanation above, review the related topic pages, then use the primary-source links below to inspect the abstract, figures, experiments, and released implementation.
Primary sources
External links are provided after the context needed to evaluate the work.
Follow-up research
Related papers and concepts
Genie 3: A General-Purpose World Model
Genie 3 generates interactive environments that can be explored in real time from text descriptions.
Cosmos 3: Omnimodal World Models for Physical AI
Cosmos 3 unifies language, image, video, audio, and action into an open world-model backbone for physical AI.
V-JEPA 2: Self-Supervised Video Models for Physical Planning
V-JEPA 2 learns predictive video representations that support visual understanding and zero-shot robot control.
GAIA-1: A Generative World Model for Autonomous Driving
GAIA-1 generates realistic driving scenarios conditioned on video, text, and vehicle actions.
Common questions
Frequently asked questions
What is the main idea of Marble: A Multimodal World Model?
Marble generates persistent 3D worlds from text, images, video, panoramas, or coarse 3D layouts.
Why is Marble: A Multimodal World Model important?
It makes the world-model layer visible to builders: generated 3D environments can become creative tools, simulation assets, and eventually training contexts.
What should I learn before reading Marble: A Multimodal World Model?
Recommended prerequisites are 3D representations, Generative models, World models.