World Labs, the startup founded by AI pioneer Fei-Fei Li, is launching its first commercial world model product. The product, named Marble, is now available via freemium and paid tiers. These tiers let users turn text prompts, photos, videos, 3D layouts, or panoramas into editable and downloadable 3D environments.
The launch of this generative world model comes a little over a year after World Labs came out of stealth with two hundred and thirty million dollars in funding. The model was first released in a limited beta preview two months ago. This launch puts the startup ahead of other competitors who are also building world models. World models are AI systems that generate an internal representation of an environment, which can then be used to predict future outcomes and plan actions.
Other startups, such as Decart and Odyssey, have released free demos of their technology. Google’s Genie model is still in a limited research preview. Marble is different from these competitors, and even from World Labs’s own real-time model called RTFM, because it creates persistent and downloadable 3D environments. It does not generate worlds on-the-fly as you explore them. The company states this approach results in less visual morphing or inconsistency. It also allows users to export their created worlds as Gaussian splats, meshes, or video files.
Marble is also the first model of its kind to offer AI-native editing tools. It includes a hybrid 3D editor that lets users block out basic spatial structures before the AI fills in the visual details. A company representative described this as a brand new category of model that generates 3D worlds, noting that the technology is something that will continue to improve over time.
Last December, World Labs demonstrated how its early models could generate interactive 3D scenes from just a single image. While impressive at the time, those scenes were somewhat cartoonish and not fully explorable, with movement limited to a small area and occasional rendering errors. In a trial of the Marble beta preview, a reviewer found it generated impressive worlds from image prompts alone, ranging from game-like environments to photorealistic versions of a living room. While scenes sometimes morphed at the edges during the beta, the company says this has been improved for the full launch.
The company is emphasizing creative control as a main theme for Marble. The goal is to provide a quick way to generate content while also allowing users deep control over their creations, ensuring the machine does not take all the creativity away from the user.
Marble offers input flexibility. While the beta only accepted single images, the full launch allows users to upload multiple images or short video clips to show a space from different angles. This enables the model to generate more realistic digital twins of real-world spaces.
A feature called Chisel is an experimental 3D editor. It lets users block out coarse spatial layouts, such as walls and boxes, and then add text prompts to guide the visual style. Marble then generates the world, effectively decoupling the structure from the style. This is similar to how HTML provides the structure of a website and CSS adds the color. Unlike text-based editing, Chisel allows for direct manipulation of objects in the scene.
Another new feature allows users to expand a world after it is generated. Users can move to an area where the world is breaking apart and command the model to generate more content in that vicinity, adding more detail. For creating extremely large spaces, users can combine multiple pre-built worlds using a composer mode.
Marble is available through four subscription tiers. The Free tier offers four generations from text, image, or panorama. The Standard tier costs twenty dollars per month for twelve generations plus multi-image and video input with advanced editing. The Pro tier is thirty-five dollars per month for twenty-five generations, scene expansion, and commercial rights. The Max tier costs ninety-five dollars per month for all features and seventy-five generations.
The company believes the initial use cases for Marble will be in gaming, visual effects for film, and virtual reality. In gaming, developers could use Marble to generate background environments and ambient spaces, then import those assets into game engines like Unity or Unreal Engine to add interactive elements and logic. It is not designed to replace the entire existing pipeline for game development, but to provide assets that can be dropped into that pipeline.
For visual effects work, the company states that Marble sidesteps the inconsistency and poor camera control that can plague AI video generators. Its 3D assets allow artists to stage scenes and control camera movements with frame-perfect precision. While World Labs is not currently focusing on virtual reality applications, the industry is described as starved for content. Marble is already compatible with the Vision Pro and Quest 3 VR headsets, and every generated world can be viewed in VR.
Marble may also have potential use cases in robotics. Training data for robotics is not as plentiful as for image and video generation. With generators like Marble, it becomes easier to simulate training environments for robots.
According to a recent manifesto by Fei-Fei Li, the CEO and co-founder of World Labs, Marble represents the first step toward creating a truly spatially intelligent world model. Li believes the next generation of world models will enable machines to achieve spatial intelligence on an entirely new level. If large language models can teach machines to read and write, systems like Marble can teach them to see and build. The ability to understand how things exist and interact in three-dimensional spaces could eventually help machines make breakthroughs beyond gaming and robotics, extending into science and medicine. Li wrote that our dreams of truly intelligent machines will not be complete without spatial intelligence.