Image model deploy
MiniTavern supports cloud image gen (OpenAI / Gemini / OpenRouter / Qwen) and local Stable Diffusion–family services:
| Local backend | App platform | Default port |
|---|---|---|
| AUTOMATIC1111 / SD WebUI | A1111(SD WebUI) | 7860 |
| ComfyUI | ComfyUI | 8188 (Desktop often 8000) |
MiniTavern on the phone does not run diffusion models on the device; images are generated on the PC (or cloud GPU). The App only sends requests and shows results.
What you need
| Item | Suggestion |
|---|---|
| PC | Windows is easiest; macOS / Linux also fine |
| GPU | NVIDIA 6GB+ is comfortable (SD 1.5); SDXL / larger models prefer 8–12GB+ |
| No discrete GPU | Try CPU (very slow) or rent a cloud GPU (hourly, not free) |
| Network | Phone and PC on the same Wi‑Fi (avoid guest isolation) |
| Disk | Models often 2–10GB+ each — leave space |
No GPU and no desire to deploy: use free cloud image tiers in Free cloud · Image gen if available, or chat only for now.
Which path
| Tool | Best for | Difficulty |
|---|---|---|
| A1111 | Classic WebUI, clear parameters, common with tavern ecosystems | Medium |
| ComfyUI | Workflows / fast new-model support; App uses a built-in default workflow | Medium |
Both end at: Configure image gen in MiniTavern.
Suggested reading order
- Pick A1111 or ComfyUI by hardware; install and download models
- Enable API + LAN listen (each guide explains)
- Connect MiniTavern and Test
- In chat use
/imagineor Attach → Image gen (see Model link · Image gen) - Back to Run local models overview
Important reminders
- On the phone use the PC’s LAN IP (e.g.
192.168.1.8) — notlocalhost/127.0.0.1. - A1111 needs
--api; opening the Web UI alone is not enough. - ComfyUI needs
--listen(or Desktop Listen0.0.0.0) for phone access. - Allow ports in the firewall; do not expose an unauthenticated image API to the public internet.
- The App’s default Comfy workflow does not consume reference images; for face lock prefer A1111 or a cloud platform that supports refs.