All projects
Mythforge Warrior Creator
Local 3D fantasy character atelier. A local LLM turns a prompt into warrior settings, Blender builds a procedural character and exports a GLB, and a Three.js preview lets you inspect and download it. No hosted inference.
Python, Ollama, Blender, Three.js, glTF / GLB
The problem
Making a game-ready character normally needs 3D modelling skills. Mythforge lets you describe or configure a fantasy warrior and get an exportable 3D model, all on your own machine.
My role
Responsibilities
Sole developer: the Python server, the prompt interpreter, the Blender generation script and the Three.js preview and export UI.
Main workflows
- 1.Pick a lineage, armor palette, weapon and colors, or type a prompt
- 2.A local LLM interprets the prompt into the supported warrior settings
- 3.Blender runs headless and builds the character mesh from those settings
- 4.The page previews the model with orbit and zoom and front, back and side views, then exports a GLB
Architecture
- Python standard-library HTTP server with no framework
- Ollama (llama3.2:1b) runs the prompt interpreter locally; the LLM chooses settings and does not generate geometry
- A Blender Python script builds and exports the procedural mesh, a Blender scene and a preview render
- Three.js frontend for the preview; character detail is limited to the parts the generator supports
Security and deployment
- Prompts are sent only to Ollama on 127.0.0.1; there is no hosted inference endpoint
- Character prompts and geometry are not uploaded by the app
- The preview currently loads Three.js and a display font from public CDNs, so it needs an internet connection until those are served locally