Fei-Fei Li’s World Labs accelerates the global model race with its first commercial product, Marble

World Labs, a startup founded by AI pioneer Fei-Fei Li, launches its first commercial world model product. Marble will be available in freemium and paid levels, allowing users to transform text prompts, photos, videos, 3D layouts, or panoramas into editable and downloadable 3D environments.

The launch of Generative World Models, which was first released in limited beta preview two months ago, comes a little more than a year after World Labs emerged from stealth with $230 million in funding, putting the startup ahead of its competitors in building world models. A world model is an AI system that generates an internal representation of the environment that can be used to predict future outcomes and plan actions.

Startups like Decart and Odyssey have released free demos, but Google’s Genie is still in limited research preview. Marble differs from these, and even from World Labs’ own real-time model, RTFM, because it creates persistent, downloadable 3D environments rather than generating worlds on the fly during exploration. The company says this reduces morphing and inconsistencies and allows users to export their worlds as Gaussian splats, meshes, or videos.

Marble is also the first of its kind to offer AI-native editing tools and a hybrid 3D editor that allows users to block out spatial structures before AI fills in visual details.

“This is a whole new category of models that generate 3D worlds, and it’s something that will get better over time. This is something that’s already gotten a lot better,” World Labs co-founder Justin Johnson told TechCrunch.

Last December, World Labs showed how an initial model could generate interactive 3D scenes based on a single image. Although impressive, the slightly cartoonish scenes were not fully explored as movement was limited to small areas and occasional rendering errors occurred.

After playing around with the beta preview, I found that Marble generates impressive worlds from just image prompts, from game-like environments to photorealistic versions of your living room. Morphing occurred at the edges of scenes, but it’s definitely improved with today’s release. That said, the worlds I generated in beta using a single prompt looked better and were more in line with my intentions than the same prompt today.

I haven’t tested the editing features yet, but Johnson says Marble will be practical for short-term gaming, VFX, and virtual reality (VR) projects.

“One of the key themes going forward for Marble is creative control,” says Johnson. “There should always be a quick path to producing something, but you should be able to dig deeper and have more control over what you’re producing. You don’t want a machine to just take the wheel and take all that creativity away from you.”

Marble’s creative control begins with input flexibility. The beta version only accepted one image, and the model had to invent invisible details in a 360-degree view. With the full release, users can now upload multiple images or short clips to view a space from different angles and have the model generate a highly realistic digital twin.

Next, use Chisel, an experimental 3D editor. It allows users to block out the rough spatial layout (think walls, boxes, or flat surfaces) and add text prompts to guide their visual style. Marble separates structure from style and generates the world in the same way that HTML provides structure for a website and CSS adds color. Unlike text-based editing, Chisel lets you manipulate objects directly.

“You just go in there, grab the 3D block that represents the couch, and move it to another location,” Johnson said.

Another new feature that gives you more editing control is the ability to expand your world.

“Once you generate a world, you can expand it once,” Johnson says. “If you move to a part of the world that’s starting to fall apart, you can basically tell the model to expand there or generate more world near where you are, so you can add more detail to that area.”

Users who want to create very large spaces can use “Composer Mode” to combine multiple worlds. Johnson showed me this using two worlds he had already built: a room made of cheese with grape chairs and a futuristic conference room in space.

The path to spatial intelligence

Marble is available in four subscription tiers. Free (4 generations from text, image, or panorama), Standard ($20/month, 12 generations plus multi-image/video input and advanced editing), Pro ($35/month, 25 generations with scene extensions and commercial rights), and Max ($95/month, all features and 75 generations).

Johnson believes Marble’s first use cases will be in games, movie visual effects, and virtual reality.

Game developers have mixed feelings about this technology. A recent Game Developers Conference survey found that one-third of respondents believe generative AI is having a negative impact on the gaming industry. This is a 12% increase from the previous year’s survey. Theft of intellectual property, energy consumption, and reduced quality due to AI-generated content were among the top concerns. And last year, a Wired investigation found that game studios like Activision Blizzard are using AI to cut corners and wage wars of attrition.

In the gaming space, Johnson sees developers using Marble to generate background environments and ambient spaces, then importing those assets into game engines like Unity and Unreal Engine to add interactive elements, logic, and code.

“This is not designed to replace the entire existing pipeline for games, it just provides assets that can be dropped into that pipeline,” he said.

For VFX work, Johnson says Marble avoids the inconsistency and poor camera control that plague AI video generators. Its 3D assets allow artists to stage scenes and control camera movement with frame-perfect precision, he said.

Johnson said WorldLabs is not currently focused on virtual reality (VR) applications, but said the industry is “hungry for content” and is excited about its launch. Marble is already compatible with Vision Pro and Quest 3 VR headsets, and all generated worlds can now be viewed in VR.

Marble may also have potential use cases in robotics. Johnson pointed out that unlike image and video generation, robotics does not have the benefit of a large repository of training data. However, generators like Marble make it easy to simulate training environments.

According to a recent manifesto from World Labs CEO and co-founder Fei-Fei Li, Marble is the first step toward building a “truly spatially intelligent world model.”

Lee believes that “the next generation of world models will enable machines to achieve a whole new level of spatial intelligence.” If large-scale language models can teach machines to read and write, Lee hopes that systems like Marble can teach machines to see and create. She says the ability to understand how things exist and interact in three-dimensional space could ultimately help machines make breakthroughs beyond gaming and robotics, and even into science and medicine.

“Our dream of truly intelligent machines is incomplete without spatial intelligence,” Lee wrote.

Do you have confidential information or documents? We report on the inner workings of the AI ​​industry, from the companies shaping the future to the people affected by their decisions. Contact Rebecca Bellan (rebecca.bellan@techcrunch.com) or Russell Brandom (russell.brandom@techcrunch.com). To communicate securely, you can contact us via Signal at @rebeccabellan.491 and russellbrandom.49.