Text to GLB with AI Free 3D model generator
Free. Use your own Anthropic or OpenAI key.
Also from Innerscene: Circadian Lab, a free lighting calculator for room illuminance, melanopic EML, and glare.
Watch how it works
How it works
- 1. Describe it or add a photoType what you want in plain words, with sizes if you know them, or add a photo as a reference. Describe an object or a person.
- 2. The AI builds and checks itThe AI writes an OpenSCAD program for an object, or sets up the character generator for a person. It looks at six rendered views of the result and fixes what is wrong, up to five builds per message.
- 3. Refine and exportThe model opens in the viewer at real size. Ask for changes in the chat, move the sliders, then download it or pick a format from the Export menu.
Why make it with AI here
- One self-contained file, which is easy to upload to a website, a three.js scene or a model-viewer tag.
- Generated people are delivered as textured GLB, so nothing is lost.
- GLB uses metres, so Android Scene Viewer and web AR place it at true size.
Prompts to try
Click one to open it in the AI panel, then change the sizes to suit.
Exporting GLB
- Download the GLB from the version list, or choose GLB in the Export menu.
- Drop it into your web page, Blender, Unity with glTFast, or Unreal through Interchange.
- On Android, open it in Scene Viewer to see it in AR.
GLB keeps colours, materials and textures in one file, in metres. File sizes stay small for objects built from simple shapes.
Best export formats: GLB, glTF
Good to know
Objects use flat colours and materials, without photo textures. Animated or rigged characters are not produced.
Cost and privacy
You use your own Anthropic or OpenAI API key. It is stored only in your browser and sent only to that provider, which bills you directly: a model typically costs $0.05 to $0.30 at Standard quality and up to about $1 at Detailed, and Innerscene charges nothing. Your key and prompts are not sent to Innerscene; we record anonymous usage statistics such as the steps used, the model chosen, token counts, a cost estimate and whether a model was made.