Generate a 3D model from a sentence
No source image needed. Describe the object and a clean, textured GLB shows up in your library · then rig it, animate it, or export it to whatever you use.
-
01
Describe the object
Name the thing, then its material, style and period. "A weathered medieval helmet with gold filigree, viking style, PBR ready" works; "a helmet" does not. Up to 500 characters.
-
02
Choose quality and variants
Standard is four tokens. High is six and returns HD textures; ultra is eight and adds detailed geometry. Ask for two or four variants when you are exploring rather than producing.
-
03
Generate and keep going
The mesh lands in your library textured. Rig it, apply a motion, retexture it, or export it in any of the five formats.
Writing a prompt that works
Be concrete about the object and its material, and leave the composition alone. These models make one object, not a scene, so "on a table, dramatic lighting" is wasted on them. Naming a material (brass, worn leather, glazed ceramic) moves the result more than any adjective about mood.
Rig-ready by default, if you ask
There is a toggle that rewrites your prompt to force a neutral T-pose, which is what automatic rigging needs to work cleanly on a character. It is free · it only changes the words sent to the engine.
Filling a scene
Generate the twenty props a level needs without sourcing twenty reference photographs first.
Concepting
Four variants of an idea, orbitable, in the time it takes to sketch one.
Placeholders that survive
Blocking geometry good enough that it does not have to be replaced before the level is playable.
How much does a text-to-3D generation cost?
Four tokens at standard quality, six at high, eight at ultra · the same ladder as an image generation on the Studio engine.
How long can the prompt be?
500 characters.
Can I generate several versions at once?
Yes, one, two or four variants per run. Each costs its own tokens.
Will the model be rigged?
Not automatically, but a humanoid result can be rigged in one step afterwards. Turn on rig-ready first and the prompt is rewritten to produce a T-pose, which makes that step work far more reliably.