Image generation
Generate images with AUTOMATIC1111, ComfyUI workflows and LoRAs, or on-device Core ML Stable Diffusion in LM Mini.
LM Mini can drive image models on your PC (A1111 / ComfyUI) or, on supported iPhones, run Core ML Stable Diffusion on-device. Personas can bind a preferred ComfyUI workflow.
Choose a backend
Settings → Image Generation
| Backend | Needs | Notes |
|---|---|---|
| AUTOMATIC1111 | A1111 WebUI API reachable | Classic txt2img, HR fix |
| ComfyUI | ComfyUI server + API workflow | Full graphs, LoRAs via placeholders |
| On-device (Core ML) | iOS build with SD assets | Private; LoRA/VAE not mirrored yet |
Master toggle: Image generation enabled. Optional review prompt before send and auto-generate on replies.
Home on Mac can proxy A1111 / ComfyUI on that Mac through Share with phone, same idea as chat.
AUTOMATIC1111
- Launch WebUI with the API enabled
- Paste the base URL into LM Mini
- Pick checkpoint, sampler, steps, CFG, size, optional HR / face restore
- Generate from the image button in chat
ComfyUI & LoRAs
- Select ComfyUI and enter the server URL
- Use the default workflow, paste API-format JSON, or pick a saved API workflow from the server
- Placeholders in string node inputs are filled at request time:
%PROMPT% %NEGATIVE% %SEED% %STEPS% %CFG% %WIDTH% %HEIGHT% %SAMPLER% %SCHEDULER% %CHECKPOINT% %LORA% %LORA_WEIGHT%
- Set LoRA name/weight in Image Generation settings when the graph expects
%LORA%
Export workflows from ComfyUI in API format, not the UI-only graph JSON.
On-device Core ML (iOS)
Download a supported checkpoint from the on-device SD models screen, select it under Image Generation → On-device, then generate without a PC. Advanced LoRA/VAE paths that need PyTorch weights are not on-device yet. macOS UI for this path may still be hidden.
Personas
Art-oriented Personas can store a default ComfyUI workflow so character chats jump into the right graph.