World Labs, the innovative startup founded by renowned AI pioneer Fei-Fei Li, has officially launched its inaugural commercial world model product named Marble. This generative AI system is now accessible through both freemium and paid subscription tiers, enabling users to transform various inputs—such as text prompts, photos, videos, 3D layouts, or panoramas—into fully editable and downloadable 3D environments.

The public launch of Marble, which was first made available in a limited beta preview two months prior, comes just over a year after World Labs emerged from stealth mode with a substantial $230 million in funding. This strategic move positions World Labs ahead of many competitors in the race to develop advanced world models. World models are sophisticated AI systems designed to construct an internal representation of an environment, allowing them to predict future outcomes and plan complex actions.

Marble distinguishes itself significantly from other offerings in the market, including free demos from startups like Decart and Odyssey, and Google’s Genie, which remains in limited research preview. Unlike these models, and even World Labs’ own real-time model, RTFM, Marble is engineered to create persistent and downloadable 3D environments. This approach means that worlds are not generated on-the-fly as a user explores, a design choice that, according to World Labs, results in markedly less morphing or inconsistency within the generated spaces. Furthermore, users gain the flexibility to export these worlds in various formats, including Gaussian splats, meshes, or videos.

A groundbreaking aspect of Marble is its introduction of AI-native editing tools and a unique hybrid 3D editor. This editor empowers users to first block out coarse spatial structures, such as walls or fundamental shapes, before the AI seamlessly fills in the intricate visual details. Justin Johnson, co-founder of World Labs, emphasized this innovation to TechCrunch, stating, "This is a brand new category of model that’s generating 3D worlds, and this is something that’s going to get better over time. It’s something we’ve already improved quite a lot."

World Labs had previously demonstrated in December how its early models could generate interactive 3D scenes from a single image. While impressive, these initial scenes were somewhat cartoonish and offered limited explorable movement, occasionally exhibiting rendering errors. Feedback from beta testers, including Johnson himself, noted that initial image prompts could generate game-like or even photorealistic environments. Johnson indicated that issues like scene morphing at the edges, observed during beta, have been significantly improved in the full commercial launch.

A core philosophy behind Marble's development is "creative control." Johnson articulated, "There should always be a quick pathway to generate something, but you should be able to dive even deeper and get a lot of control over the things that you’re generating. You don’t want the machine to just take the wheel and pull all that creativity away from you."

This emphasis on creative control is evident in Marble’s flexible input-to-output pipeline. The beta version was restricted to single image inputs, compelling the model to invent unseen details for 360-degree views. With the full launch, users can now upload multiple images or short video clips to provide the model with diverse perspectives of a space, leading to the generation of remarkably realistic digital twins.

Further enhancing user control is Chisel, an experimental 3D editor within Marble. Chisel allows users to define coarse spatial layouts—such as walls, boxes, or planes—and then apply text prompts to guide the visual style. This feature effectively decouples structure from style, akin to how HTML provides website structure and CSS adds visual styling. Unlike purely text-based editing, Chisel enables direct manipulation of objects within the 3D environment. Johnson illustrated this, explaining, "I can just go in there and grab the 3D block that represents the couch and move it somewhere else."

Another innovative feature is the ability to expand a generated world. "Once you generate a world, you can expand it up to once," Johnson noted. This means if a user approaches a boundary where the world begins to break apart, they can instruct the model to expand that region, adding more detail to the vicinity. For users aiming to create exceptionally large spaces, a "composer mode" allows for the seamless combination of multiple distinct worlds. Johnson demonstrated this by merging a whimsical room made of cheese with grape chairs and a futuristic meeting room in space.

Marble is available through four distinct subscription tiers: the Free tier offers four generations from text, image, or panorama. The Standard tier, priced at $20 per month, provides 12 generations, multi-image/video input, and advanced editing features. The Pro tier, at $35 per month, includes 25 generations, scene expansion, and commercial rights. Finally, the Max tier, at $95 per month, offers all features and 75 generations.

Johnson anticipates that the initial primary applications for Marble will span gaming, visual effects (VFX) for film, and virtual reality (VR). In the gaming industry, where a recent Game Developers Conference survey revealed a third of respondents view generative AI negatively due to concerns about IP theft, energy consumption, and quality reduction, Johnson envisions Marble assisting developers by generating background environments and ambient spaces. These assets can then be imported into game engines like Unity or Unreal Engine, where developers can add interactive elements, logic, and code. Marble is not designed to replace the entire gaming pipeline but rather to provide valuable assets that can be integrated into it.

For VFX work, Marble offers a solution to common challenges plaguing AI video generators, such as inconsistency and poor camera control. Its 3D assets empower artists to precisely stage scenes and control camera movements with frame-perfect accuracy. While World Labs is not currently prioritizing VR applications, Johnson acknowledged that the industry is "starved for content" and expressed enthusiasm for Marble's launch. The platform is already compatible with Vision Pro and Quest 3 VR headsets, allowing every generated world to be viewed in VR today.

Beyond entertainment, Marble also holds significant potential for robotics. Johnson highlighted that, unlike image and video generation, robotics lacks a vast repository of training data. Generators like Marble can address this by making it easier to simulate diverse and realistic training environments for robotic systems.

According to a manifesto by Fei-Fei Li, CEO and co-founder of World Labs, Marble represents a pivotal first step towards realizing "a truly spatially intelligent world model." Li posits that "the next generation of world models will enable machines to achieve spatial intelligence on an entirely new level." She draws a parallel to large language models teaching machines to read and write, expressing her hope that systems like Marble can empower machines to "see and build" in the three-dimensional world. This fundamental understanding of how objects exist and interact in 3D spaces could eventually lead to breakthroughs far beyond gaming and robotics, extending into fields such as science and medicine. Li concludes that "Our dreams of truly intelligent machines will not be complete without spatial intelligence."