- Text → image and image → image via
stable-diffusion.cpp(Metal-accelerated). - Works with SDXL finetunes as GGUF (RealVis, Juggernaut, DreamShaper, …) and other checkpoints — pick the active image model in Models.
- Curated styles with on-device preview thumbnails; LoRA support.
- Tune resolution, steps, seed; live progress with previews. Outputs saved to the gallery (scoped to the chat/project) and openable from chat.