Fei-Fei Li’s World Labs Raises $1 Billion from Nvidia, AMD

World Labs, a startup co-founded by renowned AI researcher Fei-Fei Li, has raised one billion dollars in a new funding round. Investors include technology giants such as AMD and NVIDIA, as well as other notable partners including Autodesk, Emerson Collective, Fidelity Management & Research Company, and Sea. With this capital, the company aims to advance its mission: developing world models that enable spatial intelligence and revolutionize areas such as storytelling, creativity, robotics, and scientific discovery.

Simultaneously with the funding announcement, World Labs has made its first product available to the general public. Marble, a generative multimodal world model, enables anyone to create spatially coherent, high-resolution, and persistent 3D worlds from images, videos, or text inputs. What was previously available only in closed beta tests is now open to all interested users.

What is World Labs?

World Labs positions itself as a research and product company at the forefront of AI development. The company was founded by Fei-Fei Li, a pioneer in artificial intelligence, together with globally recognized technologists Justin Johnson, Christoph Lassner, and Ben Mildenhall. Each brings deep expertise in machine learning, generative AI, computer vision, and graphics.

The company’s vision is to develop foundational world models that can perceive, generate, understand, and interact with the 3D world. Through spatial intelligence, the full potential of AI is to be unlocked: seeing becomes action, perception becomes thought, and imagination becomes creation. World Labs believes that spatial intelligence will enable new forms of storytelling, creativity, design, simulation, and immersive experiences in both virtual and physical worlds.

Marble: World Labs’ First Product

Marble is the centerpiece of World Labs’ current product strategy. It is a generative 3D world model that can create complete three-dimensional worlds from various inputs. The technology is based on state-of-the-art generative models and enables users to create spatially coherent and visually impressive 3D environments.

What makes Marble special is its multimodality. The system can work with a wide variety of input formats:

• Text descriptions: Users can enter simple text prompts such as “A detailed hobbit kitchen with woven baskets and copper kettles in calm pale blue daylight” and Marble generates a complete 3D world from it.
• Single images: A single photograph or AI-generated image can serve as a starting point, which Marble transforms into a navigable 3D environment.
• Multiple images: By inputting multiple views of the same scene, Marble can create more precise worlds that more closely match the user’s vision.
• Videos: Short video sequences can serve as the basis for generating 3D worlds that contain elements of real places.
• Rough 3D layouts: With the new “Chisel” tool, advanced users can sketch the structure of their world directly in 3D.

How Does the World Model Work?

The concept behind Marble is based on the idea that human perception of the world is fundamentally multimodal. We use all our senses to understand the world around us, integrating sight, hearing, touch, and speech to build a mental model of the external world. These different representations work together, enriching and reinforcing each other, enabling us to think about and act in the world.

According to World Labs, world models should function similarly. They should be massively multimodal, capable of converting whatever input signals are available into a complete 3D world, and they should be able to iteratively update their understanding of the world as new information becomes available.

Marble makes significant progress toward this vision. The system can create 3D worlds from a variety of input types and enables users to iteratively edit or expand worlds. This provides an unprecedented level of control over the creative process.

Advanced Features for Creative Control

Beyond pure generation, Marble offers extensive editing capabilities. Users can modify generated worlds after creation, whether through small local adjustments like removing an object or through more drastic interventions such as swapping objects, changing visual style, or restructuring large parts of the world.

With the “Chisel” tool, advanced users can shape the structure of their world directly in 3D. A distinction is made between structure and style: the rough 3D scene determines the structure of the world, while the text prompt controls the overall style. These two elements can be mixed in any combination, opening up an entirely new dimension of control over world generation.

Marble also enables the expansion of existing worlds. Users can select an area of a world, which Marble then fills with additional content. This can make worlds larger, add details, or improve areas that were initially rendered less sharply. Furthermore, multiple worlds can be combined into extremely large spaces, with users maintaining complete control over the arrangement.

Export and Integration into Existing Workflows

After creation, Marble worlds can be exported in various formats:

• Gaussian Splats: The highest-resolution representation for Marble worlds, representing 3D scenes as a large collection of semi-transparent particles.
• Triangle Meshes: Both low-resolution collision meshes for rough physics simulations and high-quality meshes that closely replicate the visual quality of Gaussian Splats.
• Videos: Generated worlds can be rendered as videos with pixel-perfect camera control. Marble can also “enhance” these videos by adding details, removing artifacts, and inserting dynamic elements into the scene.

These export options enable the integration of Marble worlds into many industry-standard tools and workflows, from game development to visual effects to architectural visualization.

Marble Labs: Inspiration and Use Cases

Alongside the general availability of Marble, World Labs has also launched Marble Labs, a creative hub where artists, engineers, and designers explore the boundaries of world models. The platform showcases bold ideas, real workflows, and new possibilities in areas such as gaming, visual effects, design, robotics, and more.

Marble Labs provides detailed case studies, tutorials, and documentation that give everyone the tools to learn, build, and share their own 3D worlds. From cinematic filmmaking to interactive worlds to robot simulations and therapeutic environments, these projects demonstrate how Marble is already turning imagination into reality today.

The Path to Spatial Intelligence

Despite Marble’s impressive capabilities, World Labs emphasizes that this is only one step on the path to spatial intelligence. A central future focus is interactivity. Future world models should enable both people and agents to interact with generated worlds in new ways, unlocking even more use cases in simulation, robotics, and beyond.

The billion-dollar funding gives World Labs the resources to advance this vision. With a world-class team that combines cutting-edge research with product development, and a first product that already offers impressive possibilities today, the company positions itself as a leading player in a field that could fundamentally change the way we interact with digital and physical worlds.