World Labs Launches Marble, Advancing Spatial AI
World Labs launches Marble, a generative AI system for creating stable 3D environments, marking a milestone in spatial AI.

Fei-Fei Li’s World Labs Launches Marble, Accelerating the Global Race in Spatial AI and World Models
World Labs, the AI startup founded by renowned computer scientist Fei-Fei Li, has officially launched Marble, its first commercial generative world model product, marking a significant milestone in the emerging field of spatial artificial intelligence (AI). Announced on November 12, 2025, Marble promises to transform how 3D environments are created, explored, and integrated across industries, providing a new platform for building persistent, editable, and immersive digital worlds.
What is Marble and Why It Matters
Marble is a generative AI system that creates stable 3D environments from various multimodal inputs such as text prompts, photos, videos, panoramas, and rough 3D layouts. Unlike previous transient world generators that recreate geometry on the fly, Marble produces persistent and explorable 3D scenes that maintain their spatial coherence over time. These digital worlds can be exported in multiple formats—Gaussian splats, polygonal meshes, videos, or lightweight versions optimized for real-time playback—making them compatible with leading 3D engines like Unreal, Unity, Houdini, and Blender.
This capability enables users—from game developers and filmmakers to architects and VR content creators—to generate photorealistic and interactive 3D worlds through simple prompts or visual inputs and then refine or extend these worlds seamlessly within existing digital workflows. Marble also supports native viewing on VR platforms such as Apple Vision Pro and Meta Quest 3, bridging AI-generated content with immersive display technologies.
Fei-Fei Li, who is World Labs’ CEO and co-founder, describes Marble as the first concrete step toward “spatially intelligent world models”—AI systems capable not only of rendering 3D geometry but also understanding the relationships and interactions within these environments. She compares the role of world models in AI to that of large language models in teaching machines to read and write, framing spatial intelligence as the next frontier for machine perception and creation.
Background and Company Profile
World Labs emerged from stealth just over a year ago with a $230 million funding round, assembling a team of top researchers in spatial AI, including Ben Mildenhall, Justin Johnson, and Christoph Lassner alongside Fei-Fei Li. The company quickly positioned itself at the forefront of world model research, a cutting-edge subfield of AI that focuses on building digital representations of physical spaces that AI can understand, manipulate, and generate.
The Marble product had been in limited beta since September 2025, receiving positive feedback for its ability to produce high-fidelity, editable 3D worlds from diverse inputs. Its public release marks World Labs’ transition from research phase to commercial product deployment, signaling a broader push to integrate spatial AI into creative and industrial applications.
Technological and Industry Implications
Marble’s launch highlights several important trends and impacts for the AI and tech industries:
-
Spatial AI as a New AI Paradigm: While large language models revolutionized natural language understanding and generation, spatial AI aims to endow machines with the ability to perceive and construct 3D worlds, crucial for robotics, augmented reality (AR), virtual reality (VR), and simulation.
-
Democratization of 3D Content Creation: Marble lowers the technical barrier for 3D world creation, allowing creators without deep expertise in modeling or programming to generate complex, interactive environments simply by describing or inputting visual references.
-
Cross-Platform and Immersive Integration: By enabling export to major 3D engines and direct VR headset compatibility, Marble facilitates seamless workflows from AI-generated content to immersive experiences, boosting the potential for applications in gaming, training, storytelling, and virtual collaboration.
-
Commercial Viability of World Models: Marble is one of the first products to commercialize world model research, indicating growing confidence in the market potential of spatial AI technologies and setting a competitive pace for other AI startups and tech giants investing in this space.
Future Outlook
Fei-Fei Li emphasizes that Marble represents an early but pivotal foundation for more advanced spatially intelligent AI systems. She envisions that as these technologies mature, they will unlock new AI capabilities that are currently absent—such as machines understanding spatial context, physics, and object interactions in a manner comparable to human spatial cognition.
World Labs is actively working to expand Marble’s availability and functionality, aiming to make spatial AI tools widely accessible to researchers, engineers, and creators. This development may accelerate innovation across sectors reliant on 3D content, including entertainment, architecture, robotics, and the metaverse.



