SD-GUI: Tkinter Desktop Manager for stable-diffusion.cpp

A standalone Tkinter-based desktop interface to manage, configure, and execute stable-diffusion.cpp processes — supporting text-to-image, image-to-image, video generation, inpainting, and model conversion.

Features

• Modular tabbed UI: Generator, Gallery, Execution History, and Prompt Tag Helper tabs
• Multiple generation modes: txt2img, img2img, video generation (vid_gen), inpainting, model conversion
• T5XXL & LLM text encoder support: Select separate model paths for text encoders
• VAE control: Choose custom VAE decoders, toggle VAE tiling and direct convolution
• Video generation: Flow shift and frame count settings for vid_gen mode
• Collapsible parameter groups: Organized accordion sections for img2img/hires-fix, sampler/scheduler settings, and advanced performance options
• Output preview: Live preview of the last generated image with Open File / Open Folder buttons
• Prompt Tag Helper: Categorized chip tags (art styles, lighting, camera, quality) that append to your prompt
• Execution History: SQLite-backed persistent history with double-click to reload any past run's parameters
• Gallery: Thumbnail grid of all generated outputs with file opening support
• Profile system: Save and load .env configuration presets
• Live command preview: See the full CLI command before running
• Wayland-safe clipboard: Ctrl+C/X uses wl-copy with process isolation to avoid XWayland SIGSEGV crashes
• Desktop integration: .desktop file for Linux launcher support

Requirements

1. Python 3 with tkinter library.
2. stable-diffusion.cpp built binaries in one of:
   • ~/App/stable-diffusion.cpp/build/bin/sd-cli
   • ~/App/stable-diffusion.cpp/build/bin/sd-server
3. (Linux/Wayland only) wl-clipboard package for clipboard operations:

   # Debian/Ubuntu
   sudo apt install wl-clipboard
   # Arch
   sudo pacman -S wl-clipboard
   # Fedora
   sudo dnf install wl-clipboard

Directory Structure

• src/ — Core Python files
  • app.py — Main application window and tab management
  • runner.py — Subprocess management for sd-cli/sd-server
  • styles.py — Theme, fonts, and UI styling
  • history_db.py — SQLite execution history database
  • profile_manager.py — Profile (.env) loading and saving
  • ui/ — Modular tab classes
    • tab_generator.py — Generator tab (parameters, command preview, logs)
    • tab_gallery.py — Output gallery grid
    • tab_history.py — Execution history treeview
    • tab_prompt_helper.py — Prompt tag chip selector
    • widgets.py — Reusable UI components (CollapsibleFrame, etc.)
• profiles/ — Environment configuration files (*.env) storing preset options
• models/ — Folder containing model weights (.safetensors, .gguf, .ckpt)
• output/ — Folder where generated images and videos are saved

Getting Started

1. Place models in the models/ directory (or symlink to existing models).
2. Configure profiles as .env files in the profiles/ directory.
3. Launch the interface:

   python sd-gui.py

Network Utility (ip)

The repository includes a helper script ip to quickly detect your machine's primary local network IP and generate a shareable URL for accessing a running sd-server instance.

./ip [port]

• port: Optional port number (defaults to 1234).
• -h, --help: Show usage information.

Optimal Size Configs Reference

16:9 / 9:16 (Widescreen)

• 704 x 384 (or 384 x 704)
• 896 x 512 (or 512 x 896)
• 1024 x 576 (or 576 x 1024)

2:3 / 3:2 (Portrait / Landscape)

• 448 x 640 (or 640 x 448)
• 512 x 768 (or 768 x 512)
• 640 x 960 (or 960 x 640)

1:1 (Square)

• 448 x 448
• 512 x 512
• 768 x 768

Credits & References

• stable-diffusion.cpp: A lightweight, pure C/C++ inference implementation for Stable Diffusion, developed by leejet/stable-diffusion.cpp.
• Anima Model: A 2-billion-parameter text-to-image model optimized for anime illustrations, developed by CircleStone Labs & Comfy Org. Community GGUF format quantizations are hosted by n-Arno/Anima-P3-Turbo-AIO-Q4_K.