Text to 3D

Type a description.
Get a 3D model.

Forge runs Flux image generation and TRELLIS 3D reconstruction back-to-back. Describe any object and walk away with a textured GLB you can orbit in AR, embed anywhere, or attach to your AI agent.

Open Forge → All features
Sample GLB · drag to orbit

How it works

Three steps from prompt to model

01

Write your prompt

Describe what you want — "a ceramic coffee mug with blue glaze," "a rusted iron sword," anything. Be specific about material and style for best results.

02

Flux + TRELLIS generate it

Flux renders a detailed texture image from your text. TRELLIS lifts that image into a full 3D mesh with UV-mapped textures, all in one pipeline run.

03

Download, embed, or attach

Inspect the result in the in-browser viewer, download the GLB, share a link, or pin it to your AI agent as its 3D body — all from the same page.

What you can do with it

AR inspection

Every generated model opens in WebXR AR mode with one tap — place it on your real desk or floor and walk around it.

Agent avatar

Attach the GLB to any AI agent on three.ws so the agent gets a custom 3D body instead of a default avatar.

Shareable link

Each generated model gets a permanent URL you can share — it loads in any browser without plugins or app installs.

Data flywheel

All forge generations feed the three.ws training flywheel — improving model quality for every creator over time.

Frequently asked questions

Most models complete in 30–90 seconds. Flux renders the texture image first, then TRELLIS reconstructs the mesh — the progress bar shows both stages.
GLB (binary glTF 2.0) — the industry-standard format supported by Blender, Unity, Unreal Engine, and every major 3D viewer including Apple Quick Look.
Yes. Models you generate are yours to use, modify, and ship in any project, including commercial products.
Isolated objects with clear material descriptions generate the cleanest meshes. "A ceramic mug with blue glaze" outperforms "a mug." Avoid scenes with multiple objects or complex backgrounds — those compress poorly into a single mesh.
Yes. Scan reconstructs a 3D avatar from a real photo of your face. Forge generates any object from a text description — they serve different use cases.

Start generating in 30 seconds

No account required for your first model. Type a description, watch it appear.

Open Forge → Try Scan instead