If you’ve been anywhere near tech Twitter or Bloomberg in the last 48 hours, you’ve probably seen the number: one billion dollars. That’s what [World Labs](https://www.worldlabs.ai/), the spatial intelligence startup founded by AI legend Fei-Fei Li, just pulled in for its Series C. [Bloomberg broke the story](https://www.bloomberg.com/news/articles/2026-02-18/ai-pioneer-fei-fei-li-s-startup-world-labs-raises-1-billion) on February 18th, and [TechCrunch followed up](https://techcrunch.com/2026/02/18/world-labs-lands-200m-from-autodesk-to-bring-world-models-into-3d-workflows/) with the detail that Autodesk alone chipped in $200 million. AMD, NVIDIA, Fidelity, Emerson Collective, and Sea are also on the cap table. Even [CGTN picked it up](https://news.cgtn.com/news/2026-02-19/AI-pioneer-Fei-Fei-Li-s-World-Labs-raises-1-billion-in-funding-1KT18D1b3bO/p.html), so this is very much a global story.
But the money isn’t really what caught my attention. The product is. [Marble](https://marble.worldlabs.ai/) is World Labs’ multimodal world model, and after spending some time with it, I can see why people on [Hacker News](https://news.ycombinator.com/item?id=45923544) are calling it the most impressive AI experience they’ve had in a while. You feed it a photo, a video clip, or even just a text prompt, and it spits out a spatially coherent, high-resolution 3D environment you can actually walk through, edit, and download. Not a flat panorama, not a shaky fly-through video — a persistent 3D world.
What makes Marble stand apart from the growing crowd of world generators is the editing layer. There’s an experimental tool called Chisel that lets you sketch out rough 3D layouts — walls, boxes, planes — then layer on a text prompt for visual style. Think of it like HTML for spatial structure and CSS for aesthetics, but for entire environments. You can also upload multiple images of a real space from different angles and get a surprisingly convincing digital twin back. The [World API](https://www.worldlabs.ai/blog/announcing-the-world-api) makes all of this programmable, so developers can generate worlds on-demand and plug them straight into their own apps. The [docs](https://docs.worldlabs.ai) are clean and there’s even a [quickstart repo on GitHub](https://github.com/willemhelmet/marble-api-quickstart) if you want to get your hands dirty fast.
The freemium tier is generous enough to actually try things. The [Product Hunt page](https://www.producthunt.com/products/marble-by-world-labs) has some wild examples from early users — architects doing quick spatial mockups, filmmakers pre-visualizing scenes, game designers prototyping levels in minutes instead of days. NVIDIA even published a [blog post](https://developer.nvidia.com/blog/simulate-robotic-environments-faster-with-nvidia-isaac-sim-and-world-labs-marble) about using Marble with Isaac Sim for robotics simulation.
This feels like one of those moments where a new category clicks into place. Spatial AI has been a buzzword for a couple of years now, but with Fei-Fei Li’s credibility, a billion dollars in fresh capital, and a product that actually works, World Labs is making a pretty compelling case that 3D world generation is about to become as routine as image generation already is.
Leave a comment