One Pic 3DGitHub ↗

single image → production-ready 3d

One Pic. One Click.
Full 3D.

Turn any image into a production-ready 3D model. Upload one picture — export GLB, OBJ, FBX, or STL.

Try the studio ↓Open the live gallery ↗

Studio

Drop any picture — get a real 3D model, entirely in your browser. Nothing is uploaded; export when it looks right.

Drop an image or click to browse

jpg · png · webp — processed locally, never uploaded

Demos

Generated code, running live — orbit any model, inspect its reference, read the factory that built it.

Pipeline

A staged sculpting pipeline generates and vision-reviews one pass at a time — self-correcting until every identity-defining feature clears its threshold.

  1. 01

    blockout

    Mass, proportion, silhouette — the read must survive before any detail exists.

  2. 02

    structural

    Named parts, pivots, sockets: a runtime hierarchy the code can address.

  3. 03

    form

    Bevels, rounds, panel seams — geometry that matches the reference, not approximates it.

  4. 04

    material

    Finish class and gradient stops derived from reference pixels, not memory.

  5. 05

    surface

    Micro-detail: wear, stains, engraved linework, gloss-vs-matte zones.

  6. 06

    lighting

    Controlled camera views, tone-mapping checks, per-region comparison gates.

  7. 07

    interaction

    userData.tick idle loops, articulation-ready joints, colliders.

  8. 08

    optimization

    Triangle and draw-call budgets enforced without losing identity.

Quick start

Runs under Claude Code, Codex, or OpenCode. Python 3.10+ stdlib only — nothing to install.

1 Install the skill

git clone https://github.com/img2threejs/img2threejs.git \
  ~/.claude/skills/img2threejs

2 Invoke with one image

/img2threejs Rebuild this object as a Three.js model,
keep the proportions, angles, and colours.

3 Follow the passes

The skill classifies the subject, runs the detail inventory, and gates every pass on its own — a side-by-side comparison at each step until the render matches.

FAQ

Is this photogrammetry or mesh extraction?

No. It is reconstruction-by-code: a staged pipeline writes a TypeScript THREE.Group factory from primitives, procedural shaders, and generated geometry. One reference image is enough; nothing is downloaded or scanned.

Can it do characters, not just objects?

Yes. Characters route through an anatomy-aware track — head-unit proportions, facial landmarks, pose — with an opt-in projection-first path when maximum likeness for a specific person matters. The pipeline reports per-region confidence instead of inventing hidden detail.

What do I need to run it?

An agent host (Claude Code, Codex, or OpenCode) and Python 3.10+ — the forge scripts are stdlib-only. The rendered output needs any browser that runs Three.js.

What license?

Apache License 2.0 — skill, showcase, and generated factory code alike.