Fei-Fei Li's World Labs turns single photos into playable 3D worlds
Fei-Fei Li’s startup World Labs just revealed its first project: an AI system that generates interactive, explorable 3D scenes from a single 2D photo. Users can navigate browser-rendered spaces with mouse and keyboard controls, complete with adjustable depth of field. It proves world models can maintain spatial consistency, though movement boundaries remain limited.
World Labs just showed off its first project, turning a single flat photo into an interactive, explorable 3D scene directly inside your web browser.
Founded earlier this year by AI pioneer Fei-Fei Li, the stealthy startup already has a $1 billion valuation and $230 million in the bank from big names like a16z, AMD Ventures, Intel Capital, and Eric Schmidt.
Walking into a 2D image
Turning photos into 3D models isn't entirely new, but World Labs is doing something much closer to rendering a video game level.
In a web demo you can play with right now using a mouse and keyboard, the AI generates the entire unseen environment behind the input image. You can walk around, pan the camera, and adjust a simulated depth of field slider to blur the background.
Most generative AI tools create 2D pixels, which often leads to consistency nightmares. For example, Decart's playable Minecraft generator Oasis generated a lot of buzz recently, but it quickly forgets level layouts and hallucinates errors.
World Labs claims its system maintains consistency and obeys basic physical rules. The generated spaces have actual geometry, meaning objects feel solid and keep their positions.
You can even interact with the scene on the fly, dynamically tweaking the background lighting or repainting individual objects.
That said, it is an early preview. The visuals look a bit stylized and cartoonish, navigation is restricted to a small bounded box, and objects occasionally clip or blend into each other awkwardly.
Why it matters
We think World Labs is taking the exact right bet by sidestepping the brute-force 2D video race to focus purely on persistent 3D world models.
Right now, tools like Runway and Sora produce stunning video clips, but they are effectively dynamic flat images. You cannot easily drop them into Unreal Engine, change the camera angle on take two, or let a player walk through them.
If World Labs can pull this off, game developers, filmmakers, and architects could skip weeks of tedious white-boxing and environment layout. That would turn pre-production into an afternoon project instead of an expensive studio marathon.
However, the current tech is strictly a tech demo rather than a daily driver. The invisible walls and visible rendering glitches show that hallucinations still plague the third dimension.
The real test arrives in 2025, when World Labs plans to release its first actual commercial product. Until then, it is an impressive proof of concept that proves generative AI is finally ready to step out of flatland.
Sources (1)
- techcrunch.com(opens in a new tab)Originaltechcrunch.com