mooshieui

MooshieUI is a beginner-friendly graphical interface for ComfyUI, an open-source Stable Diffusion inference backend. It abstracts ComfyUI's node-graph system behind a clean, guided UI so users can generate images without manually building or editing workflow graphs.

Available as a desktop app (Windows/Linux via Tauri) and a self-hosted web app (Docker/LAN). Source builds are supported on macOS.

Generation Modes
  • Text to Image: generate from prompts
  • Image to Image: transform an existing image with adjustable denoise
  • Inpainting: selectively repaint areas using a built-in mask editor
Notable Features
  • Smart model detection: auto-applies optimal sampler, CFG, and resolution presets for 13 model architectures (SD 1.5, SDXL, Illustrious/NoobAI, Flux, Pony Diffusion, SD3/3.5, AuraFlow, Anima/COSMOS, and more)
  • Tiled diffusion upscaling: built-in MultiDiffusion and SpotDiffusion for seamless high-resolution output
  • Compare Grid (XYZ): multi-cell parameter sweeps that auto-stitch into a labelled comparison image
  • LoRA support: unlimited LoRAs with per-LoRA strength sliders and enable/disable toggles
  • Danbooru tag autocomplete: ~100k tag database for prompt assistance
  • One-click installer: handles Python, PyTorch, and ComfyUI automatically on first launch; no manual setup required
  • 11 languages: full i18n with 860+ translation keys
Custom Nodes

MooshieUI ships custom ComfyUI nodes auto-installed at launch:

  • MooshieSoftGuidance: CFG rescale to reduce hallucinations during upscale passes
  • MooshieSmartGuidance: positive-biased adaptive guidance
  • Tiled Diffusion: MultiDiffusion and SpotDiffusion implementations
  • MooshieFaceFix: YOLOv8-based face detection and re-denoising
  • SDXL and Flux VAE adapter: for Mugen (128-channel Flux2 latent) models
  • Nanosaur support: loader, text encoder, and VAE decode for the 1.2B DiT architecture
Links

Posts