Fei-Fei Li, one of the most influential figures in artificial intelligence, has just taken a bold leap into the future of spatial computing. Her startup, World Labs, has officially launched Marble, its first commercial world model, marking a major milestone in the race to bring digital environments to life through AI.

The World Labs Marble world model is a next-generation generative AI system that can turn text prompts, images, videos, panoramas, or even 3D layouts into fully editable, downloadable 3D environments. Think of it as ChatGPT for worlds — except instead of paragraphs, it builds entire digital realities.

A World Model Designed for the Real World

Marble was first teased two months ago in a closed beta and is now rolling out globally under freemium and paid plans. The platform offers a user-friendly way to generate immersive digital environments that can be exported as Gaussian splats, meshes, or video files, all in a few clicks.

The World Labs Marble world model differs from competitors like Google’s Genie or startups such as Decart and Odyssey. Instead of creating shifting, on-the-fly scenes, Marble builds persistent, high-quality 3D worlds that stay consistent no matter how far users explore.

“This is a brand new category of model that generates 3D worlds — and it’s something that’s going to get dramatically better over time,” said Justin Johnson, co-founder of World Labs.

How Marble Changes 3D Creation Forever

What makes Marble truly stand out is its AI-native editing system and a hybrid 3D editor called Chisel. With Chisel, users can block out the basic structure of a space — like walls, floors, and furniture — and then let Marble’s AI fill in the artistic details.

“I can literally grab a 3D block that represents a couch and move it somewhere else,” Johnson explained. “It gives users real creative control — the AI doesn’t just take over your vision.”

The World Labs Marble world model also lets users expand a generated world seamlessly, adding more details or adjacent areas without breaking visual consistency. You can even stitch multiple worlds together using “composer mode,” blending surreal creations like “a room made of cheese” with a “space-themed meeting room.”

The Path to Spatial Intelligence

Marble’s launch pricing includes four tiers:

• Free: 4 generations from text, image, or panorama inputs.
• Standard ($20/month): 12 generations plus advanced editing and video input.
• Pro ($35/month): 25 generations with commercial rights and scene expansion.
• Max ($95/month): 75 generations and all features unlocked.

According to Johnson, early adopters are already using the World Labs Marble world model for gaming, visual effects (VFX), and virtual reality (VR) development. For game studios, it’s a fast way to produce environmental assets; for VFX creators, it offers precision control that AI video generators lack.

“It’s not designed to replace the entire production pipeline,” Johnson clarified, “but to accelerate creative workflows with reusable 3D assets.”

Gaming and VFX Studios Take Notice

The gaming industry’s relationship with AI remains complicated. A recent GDC survey found that 33% of developers view generative AI as a net negative for the industry. Still, Marble’s supporters believe its power lies in enhancing creativity rather than replacing it.

With World Labs Marble world model, game developers can rapidly generate background environments, import them into Unity or Unreal Engine, and then layer gameplay logic on top. Meanwhile, VFX artists can use its high-fidelity 3D assets to create precise, camera-controlled shots that AI video tools still struggle to replicate.

VR and Robotics Applications

While VR isn’t Marble’s core focus, World Labs has ensured compatibility with the Apple Vision Pro and Meta Quest 3 headsets, allowing users to explore generated environments in full virtual reality.

In robotics, Marble could become an essential training tool. Unlike traditional image or video datasets, world models can simulate realistic, navigable spaces — helping AI systems learn by interacting with virtual environments before operating in the real world.

Fei-Fei Li, who co-founded World Labs after her pioneering work at Stanford and Google AI, believes Marble is just the beginning of something bigger: teaching machines spatial intelligence.

Fei-Fei Li’s Vision for the Future

In a recent manifesto, Li wrote:

“Our dreams of truly intelligent machines will not be complete without spatial intelligence.”

She envisions World Labs Marble world model as a stepping stone toward that goal — enabling machines not only to understand language but also to perceive, build, and reason about physical space.

“If large language models can teach machines to read and write,” she continued, “then world models can teach them to see, move, and create.”

For Li, this isn’t just about gaming or art — it’s about giving AI the ability to understand how things exist and interact in 3D space, potentially transforming industries from robotics to medicine.

A New Era of Generative AI Begins

With Marble, World Labs is signaling a shift in generative AI — from text and image synthesis to spatial intelligence and simulation. As other tech giants like Google and OpenAI explore similar frontiers, Li’s team has claimed an early lead in turning imagination into navigable, editable digital worlds.

For creators, developers, and dreamers, the World Labs Marble world model might be the next big step — not just for AI, but for how we build and experience the digital universe itself.