Generates complete Godot 4 game projects from natural language descriptions using an AI pipeline. It targets developers and power users seeking rapid prototyping or automated game creation, delivering organized, functional Godot 4 projects with custom assets and code.

How It Works

This project employs two Claude Code skills—one for planning and one for execution—to orchestrate an AI pipeline. It designs game architecture, generates 2D art via Gemini and 3D models using Tripo3D, and writes all GDScript code. To ensure code quality and compensate for LLM training data gaps, it incorporates a custom GDScript language reference and lazy-loaded API documentation. Visual Quality Assurance is performed by capturing screenshots from the running Godot engine and analyzing them with Gemini Flash vision to detect and fix rendering issues like z-fighting or broken physics.

Quick Start & Requirements

• Primary install/run command: Execute ./publish.sh <project_directory> to set up a new project folder with all necessary skills.
• Prerequisites: Godot 4 (headless or editor) on PATH, Claude Code installed, Python 3 with pip, and API keys set as environment variables (GOOGLE_API_KEY for Gemini, TRIPO3D_API_KEY for Tripo3D if generating 3D assets).
• OS: Tested on Ubuntu and Debian. macOS is untested due to screenshot capture dependencies.
• Resource footprint: Generation runs can take several hours; a cloud VM with a T4 or L4 GPU is recommended for continuous operation.
• Links: Demos and prompts are mentioned but not directly linked.

Highlighted Details

• End-to-end AI pipeline for game development, from concept to runnable Godot 4 project.
• Integrated asset generation using Gemini (2D art) and Tripo3D (image-to-3D models).
• GDScript expertise enhanced by custom language reference and lazy-loaded API docs.
• Visual QA loop using Gemini Flash vision on engine screenshots to catch rendering bugs.
• Designed to run on commodity hardware, with cloud GPU acceleration recommended.

Maintenance & Community

• Roadmap includes migrating image generation to grok-imagine-image, adding Android export recipes, and publishing a full public demo.
• Progress updates can be followed via @alex_erm.
• No explicit community channels (e.g., Discord, Slack) are listed.

Licensing & Compatibility

• No license information is provided in the README.
• Compatibility for commercial use or closed-source linking is undetermined due to the lack of a specified license.

Limitations & Caveats

• macOS support is untested due to reliance on X11/xvfb/Vulkan for screenshot capture.
• Optimal results are achieved with Claude Code Opus 4.6; Sonnet 4.6 is functional but requires more user input.
• The pipeline is dependent on specific third-party LLM APIs.