← All posts

Muse Code 3D models: how to use it and what it costs

2026-11-22 · 5 min read

Muse Code, Meta's coding agent, can now generate textured 3D models from a text description or an image, and every paid plan gets it, starting at $5 a month. You ask for an object in plain English while you work, and the model file is saved into your project.

The announcement

On November 20, @MetaforDevs posted that subscribers can now "generate 3D assets in the same session where they're building a game or scene." The post says Muse Code saves a textured GLB file to your project when you ask for an object, and that you don't need a separate 3D plugin or an outside 3D subscription.

Meta's developer blog post from the same day fills in the rest. Muse Code makes a tool call to build the model, the same way it already calls Muse Image for pictures. Three models can make that call: Muse Wonder, Muse Spark and Muse Ember.

GLB is a common format for 3D on the web, so you can usually load the file in a browser scene or a game engine without converting it first.

Nano and Pro, and what each plan gets

There are two 3D models behind the feature. Nano is the faster one. Pro is the higher-quality one, and Meta keeps it for the more expensive plans.

Plan Price 3D access
Everyday Usage $5 a month Nano only, about 30 models per 5-hour window
High Usage $15 a month 5x the Everyday usage
Power Usage $50 a month 20x the Everyday usage
Maximum Usage $200 a month 80x the Everyday usage

Prices and usage multiples come from Meta's Muse Code page. The blog post says subscribers on "higher tier plans" can also use Pro and get more generations, but it doesn't say which tier Pro starts at. Check your own plan before you count on it.

There is no per-model price. Each 3D model counts against the same usage allowance as your prompts, so a long session of generating assets leaves less room for coding in that 5-hour window.

How to try it

  1. Install Muse Code. Meta's install command is curl -fsSL https://dev.meta.ai/install.sh | bash.
  2. Sign in with a paid plan. The $5 tier is enough to test Nano.
  3. Open a project and ask for an object in plain words. Meta hasn't published a slash command or a flag for this. You describe the thing, or point it at an image.

Meta's sample gallery is a fair guide to what works: a castle, a cabin, a fruit cart, a vintage pickup, a bulldozer, a goblin, a skeleton in a tutu. Meta doesn't publish the prompts behind them, so we wrote two of our own in the same spirit as a first test:

  • "Make a 3D model of a weathered wooden fruit cart and add it to the scene."
  • "Turn the photo in assets/chair.jpg into a 3D model and show it in a web viewer I can rotate."

Meta also ships two sample projects as downloads. One is a dungeon crawler mini-game with enemies, traps and spells. The other is a website comparing apple varieties in 3D. The blog pitches three uses: building games, sending models to a 3D printer, and visualizing scenes.

What it can't do

There is no built-in tool for rigging or animating characters. Meta's workaround is a prompt that asks Muse Code to rig a character with a Mesh2Motion human skeleton and Blender's automatic weights, then build a web viewer to check the result. That means the agent is scripting Blender for you, so expect it to take a few tries.

The blog also leaves out a lot. It doesn't give polygon counts, texture resolution or generation times, and it says nothing about who owns the output or whether you can sell it. The gallery needs a caveat too. Meta says every quality example was made with Pro from a single input image. On the $5 plan you get Nano, and Meta doesn't show what Nano output looks like.

Our read

We can't vouch for the 3D quality yet. The part we like is that the asset shows up where the code is. Until now, a solo developer building a small game or a 3D product page had to make the model in one tool, export it, and wire it up in another. Here the same agent makes the model and writes the code that loads it.

We'd treat the output as a draft. Without numbers on mesh density, you won't know if a model is light enough for a phone browser until you load it. For moving scenes rather than objects, Google's Genie 4 is the closer comparison, and it costs far more to get into.

What this means for businesses using AI

Custom 3D used to mean hiring a modeler. A furniture maker or a parts shop can now get a rough 3D view of a product onto a web page for the price of a coding subscription and an afternoon. A professional render will still look better, but this is cheap enough to find out whether customers care.

If you'd like help working out where tools like this fit in your business, New Face Design's free process audit is a good place to start.

08 / Start here

Which of this can you use today?

Tell us what you use today and we'll reply within one business day with what it would take. Free, 20 minutes, no pitch deck.

Email

pgorski@newfacedesign.com

Phone

+1 (773) 627-2176

Based in

Chicago area

Working with clients everywhere

Skip the form: grab a 20 minute slot.

Open the calendar →