A free, open-source UE5 plugin that generates rigged, modular game characters from concept art images — locally, privately, and with a self-improving pipeline.
Drop a concept art image into the UE5 Editor. Get back a fully rigged, UV-unwrapped, textured modular skeletal mesh — with separate swappable assets for hair, clothing, armor, weapons, and accessories — ready to animate.
Concept art image
↓
Auto-segmentation (hair / shirt / armor / boots / weapon...)
↓
Per-part 2D → 3D reconstruction (runs fully locally)
↓
Mesh cleanup + UV unwrap + texture projection
↓
Fitted to base body skeleton + skinning weights
↓
UE5 Modular Skeletal Mesh assets in your Content Browser
- Fully local pipeline — no cloud service required, no meshes or images ever leave your machine
- Your own API key — optionally use your own Claude/GPT-4V key for AI quality feedback; you control the cost
- Self-improving — the pipeline learns from corrections over time, reducing AI API calls toward zero
- Crowdsourced intelligence — opt-in to share anonymised correction patterns globally; everyone's pipeline improves
- Modular output — each component (hair, boots, sword...) is a separate UE5 Skeletal Mesh sharing one master skeleton
- Component library — browse, mix, and reuse all previously generated parts inside the Editor
- Free & open source — MIT license, always
CharacterAssetGenerator/
├── README.md
├── CONTRIBUTING.md
├── PRIVACY.md
├── LICENSE
│
├── docs/
│ ├── architecture.md ← Full pipeline architecture
│ ├── implementation_plan.md ← 7-phase build roadmap
│ ├── acceptance_criteria.md ← Pass/fail criteria for every task
│ └── privacy_policy.md ← Exactly what data is shared and when
│
├── src/
│ ├── core/ ← Core C++ library (image, IPC, DB)
│ ├── ml/ ← ONNX inference wrappers
│ ├── mesh/ ← Mesh processing (cleanup, UV, fitting)
│ ├── export/ ← FBX/glTF export layer
│ ├── server/ ← Global sync client
│ └── ue5_plugin/ ← UE5 Editor plugin (Slate UI, build targets)
│
├── python_sidecar/ ← Python ML sidecar (TripoSR, SF3D, SAM2)
│
├── server/ ← Global aggregation server (Rust/Go)
│
├── training/ ← Tier 2 classifier training pipeline (PyTorch)
│
└── tests/
├── unit/ ← Per-module unit tests
├── integration/ ← Pipeline integration tests
└── test_assets/ ← Reference images and meshes for CI
| Requirement | Version |
|---|---|
| Unreal Engine | 5.3 or later |
| CMake | 3.26+ |
| Visual Studio | 2022 (Windows) |
| Python | 3.11+ |
| CUDA (optional) | 11.8+ for GPU acceleration |
| vcpkg | latest |
# 1. Clone the repo
git clone https://github.com/YOUR_USERNAME/CharacterAssetGenerator.git
cd CharacterAssetGenerator
# 2. Install C++ dependencies
vcpkg install
# 3. Install Python sidecar dependencies
cd python_sidecar
pip install -r requirements.txt
# 4. Build
cmake -B build -DCMAKE_TOOLCHAIN_FILE=[vcpkg root]/scripts/buildsystems/vcpkg.cmake
cmake --build build --config Release
# 5. Install plugin into UE5 project
# Copy build/ue5_plugin/ into your UE5 project's Plugins/ folder
# Enable the plugin in Edit → Plugins → Character Asset GeneratorFull setup guide: docs/setup.md (coming soon)
Nothing sensitive ever leaves your machine.
What stays local: your images, generated meshes, API key, and identity.
What is optionally shared (opt-in, disabled by default): anonymised correction patterns — quality scores and settings that worked, with no geometry or imagery attached.
Full details: PRIVACY.md
See CONTRIBUTING.md for how to:
- Run the test suite
- Add a new fix type to the self-improvement engine
- Add a new component category
- Retrain the local quality classifier
See docs/implementation_plan.md for the full 7-phase, 20-week build plan with acceptance criteria per task.
MIT — see LICENSE