In a significant stride toward a new frontier in artificial intelligence, World Labs, the startup co-founded by renowned AI pioneer Fei-Fei Li, has publicly launched Marble, a first-of-its-kind generative world model.[1][2][3] The platform moves beyond the text- and image-based confines of current AI systems to generate persistent, editable, and navigable 3D environments from a wide array of inputs.[4][5] This release marks a pivotal moment in the industry's pursuit of what Li terms "spatial intelligence," aiming to imbue machines with an understanding of the physical world that mirrors human perception.[1][6] After a successful two-month beta period with a select group of creatives, Marble is now available for public use, offering a powerful new tool for creators and a glimpse into a future where AI can construct and comprehend complex, three-dimensional spaces.[1][2][3]

Marble's core strength lies in its profound multimodality, allowing users to generate rich 3D worlds from simple text prompts, single or multiple images, videos, or even coarse 3D layouts.[1][3] This flexibility provides an unprecedented level of creative control. A user can, for example, upload a few photographs of a real-world location, and Marble will intelligently stitch them together, filling in the gaps to create an immersive digital reconstruction of that space.[1][3][6] The model is not limited to replication; it can generate entirely new environments across a wide spectrum of artistic styles, from photorealistic scenes to fantastical landscapes.[7] To further empower creators, World Labs has equipped Marble with a suite of AI-native editing tools.[2] A notable feature is "Chisel," an experimental 3D sculpting mode that allows users to define the basic structure and layout of a scene before applying a desired style through text prompts, effectively decoupling form from aesthetic.[2][3][6] Users can also interactively modify their creations by removing objects, altering styles, or expanding the generated worlds, even composing multiple distinct worlds to build vast, interconnected spaces.[2][3]

The launch of Marble is the first major product release from World Labs, a company that has raised $230 million to pursue its ambitious vision.[2][8] That vision, as articulated by Fei-Fei Li, extends far beyond creating better graphics for games or films. She argues that while large language models have mastered abstract knowledge, they remain "wordsmiths in the dark," eloquent but fundamentally ungrounded in the physical reality humans navigate intuitively.[6] The development of "spatial intelligence" is presented as the next logical and necessary evolution for AI.[1][9] This concept refers to a system's ability to perceive, reason about, and interact with three-dimensional space—the kind of intelligence that allows a person to catch a set of keys tossed across a room or a firefighter to assess the structural integrity of a collapsing building.[6][9] By building models that understand geometry, physics, and the compositionality of the world, World Labs aims to create AI that can move from simply processing information to truly understanding context, a critical step for advancements in fields like robotics and embodied AI.[5][10]

The immediate implications of a tool like Marble are most apparent in creative industries. Filmmakers, game designers, and architects who participated in the beta program have already begun integrating it into their workflows, finding that tasks that once took weeks can now be accomplished in minutes.[6][7] The platform's utility is enhanced by its ability to export generated worlds into industry-standard formats.[3] Creations can be downloaded as Gaussian splats, a modern technique for rendering photorealistic 3D scenes, or as triangle meshes.[2][11] Marble can produce both high-quality visual meshes and lower-fidelity "collider" meshes intended for physics simulations, ensuring its outputs can be easily imported into game engines and other professional software.[2][3] Looking ahead, Li envisions the applications for spatial intelligence expanding dramatically. In the medium term, such technology is expected to fuel significant advances in robotics, enabling machines to navigate and manipulate objects in the real world with greater competence.[6] Further down the line, world models could revolutionize scientific discovery and healthcare by allowing for complex simulations of experiments or novel drug interactions in immersive, virtual environments.[6]

The public release of Marble represents more than just the debut of a powerful new creative application. It is a deliberate and well-funded step toward a foundational shift in artificial intelligence.[4][12] By focusing on spatial intelligence, Fei-Fei Li and World Labs are tackling a challenge that many in the field see as essential for moving beyond the limitations of current AI.[10][9] While models today can generate text and images with stunning proficiency, they lack a fundamental grounding in the three-dimensional world that shapes human and animal intelligence.[5][9] Marble, with its ability to reconstruct, generate, and simulate 3D worlds, serves as a crucial bridge between abstract data and physical space.[1] It is an early but significant move to teach machines not just to read and write, but to see and build, opening the door to a future where AI can interact with the real and virtual worlds in ways that were previously the exclusive domain of science fiction.[4][13]