Configure image gen in MiniTavern
App path: Settings → Image gen → Image gen link.
Image parameters, reply images, and keyword triggers: Model link · Image gen.
Create an image gen link
- Open MiniTavern → bottom Settings.
- Image gen → Image gen link.
- Tap New.
- Pick a platform, fill Base URL / API Key / model (varies by platform).
- Test, then save and set as default (if the list supports it).
Local A1111
First finish A1111 install with --api --listen.
| Field | Example |
|---|---|
| Platform | A1111(SD WebUI) (or custom A1111 protocol) |
| Base URL | http://YOUR_PC_IP:7860 |
| API Key | Usually empty; fill if WebUI uses --api-auth |
| Model | Empty or a WebUI checkpoint name (follow App list) |
Replace YOUR_PC_IP with the PC LAN IP.
Do not use http://127.0.0.1:7860 (that is the PC itself, not the phone).
Local ComfyUI
First finish ComfyUI install with --listen.
| Field | Example |
|---|---|
| Platform | ComfyUI (or custom ComfyUI protocol) |
| Base URL | http://YOUR_PC_IP:8188 (Desktop often :8000) |
| API Key | Usually empty |
| Model / checkpoint | Follow the list the App fetches; files must exist on the machine |
If Test fails: open the same Base URL in the phone browser to rule out firewall or missing listen.
Cloud image gen (optional)
| Platform | When |
|---|---|
| ChatGPT(OpenAI) | Official Images API (usually paid) |
| Gemini(Google) | Gemini image models (mostly paid — follow Google pricing) |
| OpenRouter | Route to image-capable upstreams |
| Qwen | Wanxiang / Qwen-Image (see Bailian free quota) |
| Custom (OpenAI Images) | SiliconFlow, Agnes, and other /images/generations gateways (image-to-image also uses generations, not /images/edits) |
Free cloud and Key tips: Free cloud · Image gen.
Image gen in chat
After a default image link is set (you also need a main chat model to analyze / rewrite prompts):
| Entry | Description |
|---|---|
Type /imagine prompt | Free image gen. The main model first decides whether the request is chat-related (e.g. “a selfie of you” → dialogue + character refs) or unrelated (e.g. “a photo of a puppy”). You can attach images in the input: the main model sees them when writing the prompt and uses them as refs; with or without attachments it still decides character/user images separately. Attachments show as small previews on the image-gen bubble. Count follows Image gen settings → Generation count. Failures leave a bubble with retry |
/imagine scene or Attach → From story | Draw the current scene from recent dialogue, fixed 1 image |
/imagine last or Attach → From last message | Draw from the last message, fixed 1 image |
/imagine background or Attach → Generate background | Environment background, fixed 1 image |
| Long-press bubble → Image gen | Draw from that bubble, fixed 1 image |
| Reply images (smart / keyword) | Enable “Reply images” in image settings and pick a trigger. After a normal message (attachments OK), side-path image gen from the chat, fixed 1 image; attachments are style refs |
With no image link, the four attachment modes or sending /imagine show “Go to settings / Cancel” — no Toast and the input is not cleared. Smart / keyword auto image gen without a link uses Toast. Enabling “Reply images” without a link also shows a modal and keeps the switch off.
Image link only, no chat model: entries that need rewrite will fail.
Quick troubleshooting
| Symptom | Check |
|---|---|
| Test fail / timeout | PC service running? --api / --listen? IP and port? Same Wi‑Fi? |
| PC browser OK, phone not | Firewall; guest isolation; used 127.0.0.1 |
| A1111 404 | Missing --api; Base URL has an extra path |
| Comfy spins forever | Comfy queue/errors on PC; OOM; no usable checkpoint |
| Refs but wrong face (Comfy) | Default workflow ignores refs → switch to A1111 or cloud |
/imagine failure bubble vanishes | Failures leave the bubble with retry; Stop still removes unfinished placeholders |
Fuller chat error codes: Error codes.