World Labs, the startup founded by AI pioneer Fei-Fei Li, has introduced its first project: an AI system that can take a single image and turn it into a video game-like 3D scene. The key difference is not just that the output has depth, but that users can move through it, interact with it, and modify parts of it.
The company describes the work as an early preview, not a finished product. Even so, it shows where a new class of generative AI tools may be headed: away from flat images and clips, and toward digital spaces that feel more stable, navigable, and useful for creative production.
What World Labs showed
The system starts with one photo. From that input, it generates the rest of the surrounding 3D scene, creating an environment that can be explored with a keyboard and mouse through a demo on World Labs' website.
World Labs described the idea simply: "[Our tech] lets you step into any image and explore it in 3D," adding that "Beyond the input image, all is generated." That means the original picture acts as the anchor, while the AI fills in the additional space needed to create a navigable world.
The scenes are rendered live in the browser. They include a controllable camera and an adjustable simulated depth of field, or DoF. When the DoF effect is increased, objects in the background appear blurrier, giving users a familiar camera-like control over how the scene is viewed.
The results are described as impressive, though somewhat cartoonish. That matters because the preview is not being presented as a final-quality production tool. It is a demonstration of interaction, depth, and scene consistency from a single image.
Why interactivity matters
Many AI systems can already convert a photo into 3D models or environments. World Labs' claim is different because its generated scenes are interactive and modifiable. Users are not only looking at a reconstructed object or a static 3D space; they are moving a camera through a generated environment.
The system can also apply interactive effects and animations. The source example includes changing the color of objects and dynamically lighting backdrops. These details point to a workflow where generated scenes are not fixed outputs, but adjustable spaces.
That distinction is important for creative fields. A still image is useful for concept work. A video can show motion. But an interactive 3D scene gives creators more control over perspective, lighting, movement, and spatial relationships.
World Labs connects this shift to larger uses in movies, games, simulators, and other digital versions of the physical world. The company wrote that most generative AI tools make 2D content like images or videos, and that generating in 3D improves control and consistency.
The role of world models
World Labs' system is part of an emerging category called "world models." These AI systems aim to simulate games and 3D environments, rather than only generate individual images, videos, or objects.
The source article contrasts World Labs' approach with other work in the same area. Some world models can simulate games and 3D spaces, but they can suffer from artifacting and consistency problems. One example given is Decart's Minecraft-simulating world model, Oasis, which has low resolution and quickly "forgets" the layout of levels.
World Labs is presented as taking a different approach. Once its scenes are generated, they remain the same. The scenes are also described as obeying the basic laws of physics, with a sense of solidity and depth.
That consistency is central to the promise of world models. If a generated space changes unpredictably, it becomes harder to use for design, games, simulation, or storytelling. A stable 3D scene gives users a stronger foundation for exploration and editing.
What still needs work
The preview has clear limits. The scenes are not fully explorable, and movement is restricted to a small area. When users try to move beyond that space, they hit a boundary.
There are also occasional rendering errors. The source article notes that objects can blend together in unnatural ways. These issues show that the system is still early, especially if the goal is to support professional creative workflows.
World Labs has acknowledged that the preview is early. The company said it is working to improve the size and fidelity of its generated worlds, while also testing new ways for users to interact with them.
The current limitations do not erase the significance of the demonstration. Instead, they define the gap between a compelling browser demo and a product that could be used in production by artists, designers, developers, filmmakers, engineers, video game developers, or movie studios.
Where World Labs is headed
World Labs launched earlier this year and has raised $230 million in venture capital. Its investors include Andreessen Horowitz (a16z), Ashton Kutcher, Intel Capital, AMD Ventures, and Eric Schmidt. The company is valued at over $1 billion and hopes to have its first product ready in 2025.
The company's ambitions extend beyond interactive scenes. It plans to build tools for professionals such as artists, designers, developers, filmmakers, and engineers. Its target customers range from video game developers to movie studios.
World Labs co-founder Justin Johnson described the opportunity on a recent episode of the a16z podcast. He said that creating virtual, interactive worlds is already possible, but that it costs hundreds and hundreds of millions of dollars and a ton of development time. His view is that world models could produce not just an image or clip, but a fully simulated, vibrant, and interactive 3D world.
For now, the most important takeaway is narrower and more concrete: World Labs has shown an AI system that can generate a stable, interactive 3D scene from a single photo. It is limited, imperfect, and early. But it also shows how generative AI may move from making media to creating spaces that people can enter, adjust, and build on.