Quick Start¶
This guide gets you from zero to generating your first 3D asset in about 5 minutes.
1. Install assgen¶
Verify:
2. Start the server (optional)¶
The client auto-starts a local server on first use, so this step is optional. If you want the server running in the background from the start:
3. Check what tasks and models are available¶
This prints a full tree of every game-dev task alongside its assigned AI model:
🎮 assgen — Game Asset Generation Pipeline
├── 🖼 Visual Assets
│ ├── 📐 3D Model Generation
│ │ ├── visual.model.create [stabilityai/TripoSR] Generate a 3D mesh...
│ │ ├── visual.model.splat [huggingface/gaussian-...] Gaussian Splatting...
...
Filter by domain:
4. Generate your first asset¶
The fastest way to get a complete, game-ready asset is a compose command. One prompt, one command — the pipeline handles everything:
# Complete static prop: concept → mesh → UV → texture → LODs → collider → export
assgen compose prop "wooden barrel, medieval dungeon" --engine godot --wait
# Full PBR material set: albedo + normal + roughness + AO + metallic
assgen compose material "mossy cobblestone, wet" --wait
# Level audio suite: ambient loop + music + 5 SFX
assgen compose soundscape "dungeon ambience, distant dripping" --wait
See the Compose Pipelines guide for all commands.
For finer control, run each step manually:
5. Track jobs¶
# List recent jobs
assgen jobs list
# Check status (use the first 8 characters of the job ID)
assgen jobs status a1b2c3d4
# Wait for a job with a live progress bar
assgen jobs wait a1b2c3d4
6. Configure a model override¶
Each task uses a default model from the catalog. Override it for any task:
# Search HuggingFace interactively
assgen config set visual.model.create
# Or specify directly
assgen config set visual.model.create --model-id stabilityai/TripoSR
# Revert to built-in default
assgen config remove visual.model.create
7. Point at a remote GPU server (optional)¶
If you have a desktop with a GPU, run assgen-server there and point your laptop at it.
On the GPU machine (Linux or Windows):
# Bind to all interfaces so LAN clients can connect
assgen server config set host "0.0.0.0"
# Start the server
assgen-server start --daemon
On the client machine (laptop):
# Point at the GPU machine (replace with its LAN IP)
assgen client config set-server http://192.168.1.100:8432
# Verify connectivity
assgen client config show # should show "healthy"
If the health check fails, ensure:
- The server is running (
assgen server statuson the GPU machine) - Port 8432 is open (Windows Firewall / UFW — see Server Setup)
- Both machines are on the same network
Pre-download recommended models on the server to avoid long waits on first use:
What's next?¶
- Compose Pipelines — one command, complete asset pipelines
- CLI Reference — every command and option
- Configuration — server, client, and model catalog settings
- Model Validation — allow-lists and HF pipeline-tag checks
- Server Setup — systemd service, firewall, remote access