Luma AI drops text-to-3D generator Genie on Discord

Anyone who has opened Blender knows that pulling vertices and sculpting even a decent-looking teapot can eat up an entire afternoon.

Now, NeRF startup Luma AI is trying to bring the Midjourney moment to 3D. The company just launched a free research preview of Genie, a text-to-3D foundation model running directly inside Discord.

Four models in 20 seconds

The workflow feels instantly familiar if you have ever used Midjourney.

You type a text prompt into the Discord bot, and within roughly 20 seconds, Genie hands you four simple 3D models.

Each generation includes an interactive preview link where you can rotate the object in full 3D, alongside an option to download the asset as a standard glb file.

If the initial draft looks promising, you can trigger a refine pass. That high-resolution refinement takes about 20 minutes, generating much tighter meshes and cleaner materials.

From smartphone NeRFs to 3D foundation models

Luma AI made its name turning phone videos into interactive NeRFs, but Genie marks its first big step into native generative 3D.

The space has been heating up since 2021 with projects like Google's Dreamfusion, OpenAI's Point-E and Shap-E, and Nvidia's Magic3D. Most of those tools, however, remained trapped in research repos and complex setups.

Luma is explicitly targeting indie game developers and virtual production studios. Instead of building every background asset from scratch, artists can rough out ideas with text prompts.

Genie is completely free during its current research preview, though the team noted it will eventually become a paid service once the foundation model matures.

Why it matters

We think Luma's smartest move here is distribution rather than pure rendering fidelity.

For indie developers and concept artists, generating four downloadable glb meshes in 20 seconds is an incredible prototyping shortcut. You can drop them straight into Unreal or Unity for quick level greyboxing without touching a sculpting tool.

That said, it is not production-ready yet. The meshes from rapid generations will almost certainly require manual retopology before rigging or animation, and waiting 20 minutes for a refined version breaks creative flow.

Still, as 2D image generation matures into a crowded market, 3D foundation models are clearly the next major battleground. Luma just proved that text-to-3D can be fast, interactive, and accessible to anyone with a Discord account.