Image To 3D Kaggle Skill — Development skill for Claude Code

Kaggle Asset Factory — a Claude Code skill

Turn concept images into textured, game-ready 3D models (GLB) **for free**, using the weekly GPU quota of your own Kaggle account. Claude Code pushes each batch as a private Kaggle script, waits for it, downloads the models and makes them Unity / three.js ready with Blender. Every run stops itself, so no GPU time is wasted on idle notebooks.

Today it runs [Hunyuan3D-2](https://github.com/Tencent-Hunyuan/Hunyuan3D-2) (turbo shape + turbo paint) for image-to-3D. On Kaggle's 2x T4 a batch of ten images takes about 12 GPU minutes.

concept.png ──► Kaggle GPU (Hunyuan3D-2) ──► concept.glb (textured) ──► Blender ──► concept_unity.glb + concept_preview.png

Install

git clone https://github.com/otmanebaraka/image-to-3d-kaggle-skill && cp -r image-to-3d-kaggle-skill/kaggle-asset-factory ~/.claude/skills/

(Windows PowerShell: `Copy-Item -Recurse image-to-3d-kaggle-skill\kaggle-asset-factory $HOME\.claude\skills\`.) For one project only, copy it to `/.claude/skills/` instead.

Setup (once)

1. **Kaggle account** with a verified phone number (Kaggle requires it for GPUs and internet access in scripts).

2. **Kaggle CLI:** `pip install kaggle` (version 1.7+; the quota helper uses the `kagglesdk` it ships with).

3. **Your own API token:** kaggle.com → *Settings* → *API* → *Create New Token*. Save it to **one** of:

   • ~/.kaggle/access_token (new-style token), or
   • ~/.kaggle/kaggle.json ({"username": "...", "key": "..."}, legacy), or
   • the environment variables KAGGLE_USERNAME and KAGGLE_KEY.

   Keep it private: never commit it, paste it in chat or put it in a project folder. `kaggle config view` should now print your username.

4. **Blender 4.x or 5.x** (optional, for `_unity.glb` and previews): on PATH as `blender`, or set `BLENDER` to its full path.

5. **Build the cache** (once, about 4 GPU minutes; cuts every later run by ~3.5 min): `python ~/.claude/skills/kaggle-asset