Meta has launched a new AssetGen system that can generate full-blown 3D virtual worlds for VR using just a single text prompt, as part of its ongoing work on generative AI for the metaverse.

If you’ve ever had an idea for a virtual world but didn’t know how to build it, Meta’s AI might soon make that as easy as writing it down.

This tech doesn’t just churn out buildings or props—it’s designing entire VR-ready environments that look handcrafted and run on Meta’s platforms in real time.

Meta announced the new AssetGen model during its presentation at the IEEE VR Conference, as part of a broader update to its Generative AI for Virtual Reality (GAI4VR) research.

The system turns plain language prompts into complex 3D environments built with physically based rendering (PBR) materials, lighting, and spatial layout, all optimized for real-time use in platforms like Horizon Worlds or Quest apps.

Let’s say you type something like “a futuristic Tokyo street at night”—AssetGen can now generate the full scene: neon-lit storefronts, dynamic reflections, and even optimized geometry that won’t break your GPU.

Meta’s team emphasized that the environments are not just static assets.

These worlds come layered with metadata and structural context, making them usable in actual game engines or VR creative tools.

They’re optimized with occlusion-aware layout logic, meaning everything you see is positioned not just for aesthetic purposes, but also for performance.

That’s a big deal for VR, where frame rate is king and clunky assets kill immersion fast.

What makes this stand out is the leap in speed and accessibility. Traditionally, building a virtual world takes hours or days, sometimes weeks, even with professional tools like Unity or Unreal.

With AssetGen, you’re looking at minutes from text prompt to working scene.

And this is just part of a larger play.

Meta is betting on generative AI as the glue that makes the metaverse less of a buzzword and more of a usable space.

The company has already released tools like Emu for image generation and Make-A-Scene for creative composition, but AssetGen pushes things a step further—it’s about building experiences you can step into.

Developers, creators, and digital architects can also remix and iterate on the generated scenes.

Meta says they plan to open-source parts of the pipeline to encourage innovation outside its ecosystem.

While AssetGen is still a research project, it gives a clear look at Meta’s ambition: to collapse the barriers between imagination and creation in VR.