World Labs Launches Marble: Revolutionary AI Creates Explorable 3D Worlds from Text

The era of spending months building game environments with teams of 3D artists might be ending. Marble generates complete three-dimensional worlds from simple text, images, or video — and delivers them ready for download as Gaussian splats, meshes, or video files.

This isn't another web app that spits out pretty pictures. We're talking about spatial intelligence — the ability for machines to understand how objects exist and interact in three-dimensional space.

🎮 World Models: From Research Labs to Reality

World models aren't new. Google's Genie, Decart, and Odyssey have been experimenting for years. But there's a critical problem: most systems generate worlds "on-the-fly" as you explore them.

The result? Distortions at the edges of your vision. Inconsistencies when you turn your head. Environments that look stunning for 30 seconds before they start to "melt".

Justin Johnson, World Labs co-founder, puts it simply: "This is a brand new category of models that creates 3D worlds." The difference between Marble and the rest is that you can take these worlds and use them in Unreal Engine, Unity, or any other game engine.

From Living Room Photo to Playable Game

In the beta version tested by TechCrunch, Marble managed to create photorealistic representations of a house from a single photograph. Impressive, but problematic — the system had to "guess" what was hidden beyond the frame.

The final version supports multiple images and soon videos. You can upload photos from different angles of the same space and Marble creates a reasonably accurate digital twin.

But the real innovation is called Chisel — an experimental 3D editing tool. Instead of describing the world you want with words, you build rough layouts — walls, boxes, planes — then add text prompts for styling.

⚡ Chisel: HTML for 3D Worlds

The analogy World Labs uses is revealing. Chisel builds the structure like HTML, and text prompts add styling like CSS.

"I can grab the 3D box representing the couch and move it elsewhere," Johnson explains. It's the difference between asking an AI to change something by describing it with words, and changing it directly with your hands.

There's also world expansion capability. When you reach the edges of a generated environment and it starts to "break," you can ask Marble to generate more world in that area.

For extravagant projects, "composer mode" lets you combine multiple worlds. Johnson demonstrated this by merging a room made of cheese with grape chairs and a futuristic conference hall in space. Why not?

🎬 Gaming, VFX and Industry Pushback

World Labs targets three main markets: gaming, visual effects for film, and virtual reality. But the reception from the gaming industry isn't enthusiastic.

According to a recent Game Developers Conference survey, one-third of developers believe generative AI has a negative impact on the industry. The concerns? Intellectual property theft, energy consumption, and quality degradation from AI-generated content.

Johnson sees it differently. This won't replace entire development teams. It'll simply provide assets that can be imported into existing pipelines — background environments, ambient spaces that get filled with interactive elements later.

VFX: The End of "AI Video Morphing"

In visual effects, Marble promises to solve a chronic problem with AI video generators. Instead of messy camera control and constant morphing, 3D assets allow frame-perfect precision in camera movements.

Think about producing a film where you need 20 different angles of the same environment. Traditionally, you either build a physical set or spend weeks on 3D modeling. With Marble, a reference photo can become a full 3D world in minutes.

🚀 Spatial Intelligence: Li's Big Bet

Behind Marble lies a bigger vision. Fei-Fei Li — the woman who created ImageNet in 2009 and laid the foundation for modern computer vision — talks about "spatial intelligence" as the next leap in artificial intelligence.

"If large language models teach machines to read and write, systems like Marble can teach them to see and build."

— Fei-Fei Li, CEO World Labs

The idea is that understanding three-dimensional space isn't just useful for gaming and movies. It's the foundation for truly intelligent machines — robots that understand their world, autonomous vehicles that "see" beyond cameras and sensors, even applications in science and medicine.

Robotics: The Secret Dream

Here's where the big opportunity hides. Robotics doesn't have the luxury of massive datasets like image processing or text. With generators like Marble, simulating training environments for robots becomes easier.

Imagine a world where a robot can train in thousands of different generated environments before touching the real world. What Li calls "Real2Sim transfer" — translating experiences from the real to the virtual world and back.

💰 Pricing and Accessibility

Marble launches with four pricing tiers that show the company's targeting. The Free plan includes 4 generations from text, image, or panorama. Enough for experimentation, too little for serious work.

Standard at roughly $20 per month gives 12 generations, multi-image/video input, and advanced editing capabilities. Pro at $35 per month adds scene expansion and commercial usage rights.

The Max plan at $95 per month with 75 generations shows where World Labs is aiming — studios and companies that want to integrate Marble into their production pipelines.

The critical question is whether the technology is mature enough for professional use. Early reviews report impressive results, but also occasional issues with consistency at the edges of generated environments.

🔬 The Technology Behind Marble

Here's where things get technical. Marble isn't just a scaled-up image generator. It's a spatial reasoning system that understands geometry, physics, and optical relationships between objects.

The system creates 3D representations that can be exported to various formats — Gaussian splats for high-quality rendering, meshes for game engines, or simple videos for quick preview. This output flexibility is critical for industry adoption.

The neural architecture includes mechanisms for spatial consistency — something missing from most AI video models of 2026. When you walk around an object in the generated world, you see it from all angles without it changing or "melting".

Comparison with Competitors

Google Genie remains in research preview. NVIDIA Cosmos focuses more on robotics simulations. Decart has impressive demos but no commercial product.

World Labs is trying to close the gap between research and actual use. The fact that Marble is already available for purchase and download is a significant advantage.

🎯 The Future of 3D Worlds

3D content creation is changing fast. Marble is the first step, not the last.

Next year we expect integration with larger 3D authoring tools, better compatibility with game engines, and possibly real-time collaboration features. World Labs has already shown case studies with Unreal Engine and Unity integration.

But the biggest question isn't technical — it's economic. How quickly will AI-generated assets replace traditional 3D production? And what will that mean for the thousands of 3D artists worldwide?

Fei-Fei Li has seen this movie before. The ImageNet she created in 2009 brought a revolution to computer vision — but also created new opportunities for those who adapted. The question is whether the 3D content industry is ready for the next revolution.