Skip to content

assgen

AI-driven game asset generation pipeline — a Kubernetes-style client/server CLI that lets you generate 3D models, textures, rigs, animations, sound effects, and music using open-source HuggingFace models running locally on your GPU.

CI Latest Release Python 3.11+ MIT License

What is assgen?

assgen is a command-line tool designed for indie game developers and artists who want to leverage AI for asset creation — without cloud costs, without proprietary APIs, and without leaving their terminal.

You describe what you want. assgen dispatches the work to a local server that downloads the right model from HuggingFace, runs inference on your GPU, and saves the result.

# Generate a 3D prop
assgen gen visual model create --prompt "low-poly medieval sword" --wait

# Generate a sound effect
assgen gen audio sfx generate "laser gun firing" --wait

# View all available tasks and their assigned AI models
assgen tasks

Architecture

┌─────────────────────────────────────────────────────┐
│  assgen  (client CLI, runs anywhere)                 │
└──────────────────────┬──────────────────────────────┘
                       │ HTTP REST
┌──────────────────────▼──────────────────────────────┐
│  assgen-server  (FastAPI + SQLite job queue)         │
│  ├── ModelManager   (HuggingFace Hub download/cache) │
│  ├── WorkerThread   (GPU inference)                  │
│  └── Validation     (allow-list + HF tag checking)  │
└─────────────────────────────────────────────────────┘

The client auto-starts a local server on first use, or you can point it at a remote GPU machine:

assgen client config set-server http://my-gpu-desktop:8432

Feature Highlights

  • Text/image → rigged 3D mesh (.glb, .fbx, .obj)
  • Auto UV unwrapping and PBR texture generation
  • Gaussian Splatting (3DGS) from multi-view images
  • Auto-rigging and skin weight generation (UniRig)
  • Animation: keyframe generation, mocap from video
  • Sound effects from text prompts (AudioLDM2)
  • Music composition (MusicGen)
  • Voice/TTS and NPC dialog batching (Bark)
  • Scene lighting: HDRI sky generation, GI lightmap baking
  • Physics: collision mesh generation, cloth sim baking
  • Multi-step workflows with batch processing
  • Engine export (Unity, Unreal, Godot)
  • Server allow_list: restrict which HF models can be downloaded
  • Pipeline-tag validation: prevents mismatched models (e.g. TTS for 3D generation)
  • Configurable per-server overrides

Requirements

  • Python 3.11+
  • For inference: an NVIDIA GPU (RTX 4070 or better recommended — 12GB VRAM handles SDXL, TripoSR, AudioLDM2)
  • For check-only / orchestration: CPU is fine

Workflow Guides

Jump straight to a guide written for your role:

Role Engine Guide
Solo indie dev Godot Prop → texture → music in a lunch break
Environment / level artist Unreal Engine 5 Biome textures, HDRI skies, seam inpainting
Character artist Unity Concept → mesh → rig → animate
Game audio designer Any SFX variants, adaptive music, VO scratch
Pipeline TD / technical artist Any (CI/CD) GitHub Actions, allow-list, QA validation

Get started → View source on GitHub →