Generating 3A Games by Delegating to a Coding Agent
It provides the skill definitions and pipeline glue so coding agents can turn a text game brief into engine-ready assets and runtime code across UE5, Unity, Godot, and more.

What it does
3AGameFactory is an open-source orchestration layer that treats game development as an agentic workflow. A coding agent—Codex, Claude Code, or similar—reads structured skill definitions to generate images, 3D meshes, motion clips, audio, CG video, and engine-native gameplay code. It targets UE5, Blender, Unity, Godot 4, and three.js, aiming to turn a design brief into a coordinated pipeline of assets and runtime logic.
The interesting bit
The project’s real trick is turning documentation into a control plane. agent_skills/setting_overview.md acts as the agent’s routing layer, telling a generalist LLM whether to invoke Meshy for a 3D weapon or to generate Godot UI boilerplate. It is essentially an attempt to automate the integration decisions that usually require a human technical director.
Key highlights
- Supports end-to-end asset generation across images, 3D objects, scenes, motion, audio, and CG video, using local models like MiniMax H3 or cloud APIs such as Seedance.
- Provides engine-specific adapters and agent context for UE5, Unity, Godot 4, Blender, and three.js, generating both assets and engine-native mechanics/UI code.
- Demos include playable fighting, FPS, and racing games where AI-generated assets (via Meshy, Hunyuan3D) are mixed with open-source libraries like Mixamo and Unity asset store scenes.
- Separates the game-generation path (agent-driven) from the contributor path, which includes a CPU-only smoke harness for testing model wrappers and operators without a GPU.
- Organizes generated artifacts by game, run, task kind, and task ID under
test_data/outputs/, suggesting a batch-oriented, reproducible pipeline structure.
Caveats
- The framework is not a standalone executable; it is a context-and-skill library that requires an external coding agent to read the skills and drive execution.
- Generated demos currently rely on a mix of AI-generated assets and downloaded or open-source content—such as Mixamo motion and Unity asset library scenes—so it is not a fully closed-loop generative pipeline.
- Local CG-video generation is currently demonstrated at 720p using MiniMax H3, with the README noting that higher fidelity requires costlier cloud APIs.
Verdict
Worth exploring if you are building agentic workflows for creative tooling or want a structured starting point for automated asset pipelines. If you are looking for a fully self-contained, one-click game generator that runs without external agents or third-party asset libraries, this is not it yet.
Frequently asked
- What is OpenDCAI/GameFactory-3A?
- It provides the skill definitions and pipeline glue so coding agents can turn a text game brief into engine-ready assets and runtime code across UE5, Unity, Godot, and more.
- Is GameFactory-3A open source?
- Yes — OpenDCAI/GameFactory-3A is open source, released under the Apache-2.0 license.
- What language is GameFactory-3A written in?
- OpenDCAI/GameFactory-3A is primarily written in Python.
- How popular is GameFactory-3A?
- OpenDCAI/GameFactory-3A has 625 stars on GitHub.
- Where can I find GameFactory-3A?
- OpenDCAI/GameFactory-3A is on GitHub at https://github.com/OpenDCAI/GameFactory-3A.