Meta's new 3D Gen turns text to 3D in under a minute

Creating a textured 3D asset from scratch usually takes an artist days of tedious polygon manipulation and UV unwrapping.

Meta wants to compress that entire pipeline into a single coffee break.

Meta's GenAI research team just introduced Meta 3D Gen, a new combined AI system capable of generating high-fidelity 3D assets from text prompts in under a minute.

Thirty seconds for geometry, twenty for texture

Most existing 3D generative models struggle with muddy textures, weird geometry, or painfully slow rendering pipelines.

Meta tackles this by splitting the problem into two specialized stages.

First, a component called Meta 3D AssetGen creates the initial 3D mesh and basic materials from your prompt, clocking an average inference time of around 30 seconds.

Next, the asset moves to Meta 3D TextureGen, which refines or completely repaints the surface with high-resolution details in about 20 seconds.

According to the researchers, the pipeline simultaneously represents 3D objects across view space, volumetric space, and UV texture space.

Critically, the system supports physically-based rendering material maps, meaning the resulting models interact realistically with game engine lighting out of the box.

It even works backward: users can feed in human-made 3D meshes and use text prompts to generate new custom skins and textures on demand.

When tested against industry baselines, human evaluators preferred Meta's two-stage outputs 68% of the time over single-stage alternatives.

The catch? The project is currently still a research preview and not yet open to the public.

Why it matters

Meta's decision to decouple mesh generation from texturing is the smartest move here.

In practical game development and VFX workflows, tweaking materials happens infinitely more often than remaking core geometry. By giving TextureGen the ability to re-skin existing, artist-crafted models via text prompts, Meta is building a tool that actually fits into real-world production pipelines rather than just serving as a novelty demo.

For indie studios and virtual reality developers, this kind of speed could drastically cut down the cost of building large-scale background environments.

That said, the biggest question mark remains mesh topology. High-resolution texture maps look great in renders, but game engines demand clean edge loops and optimized polygon budgets. If the underlying wireframe is a chaotic mess, artists will still spend hours cleaning it up manually.

Still, Meta clearly understands where its own bottlenecks lie: if you want people spending time in VR headsets, you need millions of 3D objects, and you need them fast.