Meta’s WorldGen AI system generates trimesh 3D worlds from text prompts, though the company doesn’t think it’s ready for Horizon Worlds yet.
Meta first teased that its Horizon Worlds creation tools would get the ability to AI-generate entire 3D worlds back in May, when announcing the related AssetGen 2.0 model. Then, in June, the company revealed that this feature would be called Environment Generation, teased example generations, and said it would launch “very soon”.
David Heaney
Environment Generation launched in August, but it was (and remains) only capable of generating a very specific kind of island, a very limited scope compared to the goal of generic world creation.
What Is Horizon Worlds Desktop Editor?
Horizon Worlds Desktop Editor is a flatscreen Windows PC application Meta released in early access in February, alongside deprecating the in-VR creation tools of Horizon Worlds.
The editor offers the ability to import 3D assets, images, and sound files, place them in a 3D landscape, and implement game logic and other functionality using TypeScript, a popular offshoot of JavaScript. These worlds are then immediately playable and multiplayer-capable in Horizon Worlds.
In the US, UK, Canada, EU, Australia, and New Zealand, creators can also AI-generate 3D meshes, textures, skyboxes, sound effects, ambient audio, and TypeScript.
You can download Horizon Worlds Desktop Editor here.
At Connect 2025 in September, Meta teased an overhaul of its Horizon Worlds creation tools, called Horizon Studio, which hasn’t yet launched. The tease depicted an AI Assistant capable of generating just about anything a creator wants, including entire worlds, specific assets, custom NPCs, and specific gameplay mechanics, in a matter of seconds or minutes. But it’s unclear whether what Meta was showing was notional or representative of real technology it was waiting to deploy.
David Heaney
That brings us to WorldGen, the new AI system Meta published a paper for.
Meta describes it as “a state-of-the-art end-to-end system for generating interactive and navigable 3D worlds from a single text prompt”, leveraging a chain of 2D and 3D techniques, rather than being a single model.
“WorldGen is built on a combination of procedural reasoning, diffusion-based 3D generation, and object-aware scene decomposition. The result is geometrically consistent, visually rich, and render-efficient 3D worlds for gaming, simulation, and immersive social environments.”
To be clear, this is not producing a Gaussian splat like World Labs’ Marble, nor an interactive video stream like Google DeepMind’s Genie 3.
Meta’s WorldGen creates a layout of traditional trimesh 3D assets, making it fully compatible with traditional game engines and rendering pipelines. And it also includes a navmesh for collision detection and NPC traversal.
0:00
Here’s the underlying sequence WorldGen goes through after you input a prompt, according to Meta:
(1) Planning
1. Procedural blockout generation
2. Navmesh extraction
3. Reference image generation(2) Reconstruction
1. Image-to-3D base model
2. Navmesh-based scene generation
3. Initial scene texture generation
(3) Decomposition
1. Part extraction with accelerated AutoPartGen for scenes
2. Data curation for scene decomposition(4) Refinement
1. Image enhancement
2. Mesh refinement model
3. Texturing model
So why isn’t WorldGen rolling out in Horizon Worlds Desktop Editor, or at least being announced as a launch feature for Horizon Studio?
0:00
0:00
0:00
0:00
Meta says it’s not satisfied with the fact that WorldGen currently only produces 50×50 meter spaces, and that it takes a long time to do so. The company says it’s working to address both limitations.
It seems like a greatly upgraded future version of WorldGen will be necessary to deliver on the promise of Horizon Studio that Meta teased at Connect, and given the rate of advancement in AI, it’s very possible that the company will be able to achieve exactly that sometime in 2026.
