Skip to content

README.md

Latest commit

 

History

History
148 lines (115 loc) · 6.66 KB

README.md

File metadata and controls

148 lines (115 loc) · 6.66 KB
Raw

🧪 LoRA x Sampler x Scheduler Grid (Forge)

💡 An advanced script for Stable Diffusion Forge that generates comprehensive comparison grids exploring LoRA models across different samplers, schedulers, and weight combinations.

📝 Description

This script creates detailed image grids in two operational modes: XY Grid for systematic combinatorial testing and Batch Grid for custom sampler-scheduler pair configurations. It's designed to help artists and researchers visualize the impact of different LoRA weights across various generation parameters.

✨ Features

  • 🎯 Dual Operation Modes: XY Grid for automatic matrix generation and Batch Grid for customized pair combinations
  • 🔄 LoRA Weight Testing: Flexible weight range configuration with min/max values and step precision
  • 🏷️ Smart Label Positioning: Configurable label placement (Top, Bottom, Left, Right) with automatic collision detection
  • ⚡ Trigger Word Support: Optional trigger word integration with display toggle
  • 🛡️ Error Resilience: Fallback image generation with detailed error information when generation fails
  • 📊 Progress Monitoring: Real-time generation progress logging with detailed parameter information
  • 🖼️ Flexible Output: Individual cell saving and multiple format support (WEBP/PNG)
  • 🎨 Custom Font Support: Barlow-SemiBold.ttf integration for improved label readability
  • 📏 Auto-Scaling: Automatic downscaling of large grids beyond 16380px dimensions
  • 🛑 Process Control: Dedicated stop generation button for long-running operations

⚙️ Installation

❗ Important: This is a script, not an extension.
Forge only loads .py files placed directly inside the scripts/ directory. After cloning, you must move the main .py file out of the cloned folder and place it into /scripts/.

  1. Clone the repository into your Forge /scripts/ folder: git clone https://github.com/PupaBoo/lora_x_sampler_x_scheduler_grid_forge

  2. Navigate to the project directory: cd Forge-Grid-Sampler-Scheduler

  3. Install dependencies: pip install -r requirements.txt

  4. Move the script file lora_x_sampler_x_scheduler_grid(forge).py from the cloned directory directly into your Forge /scripts/ folder.

  5. (Optional) Place the Barlow-SemiBold.ttf font in the scripts/ folder for improved label display.

  6. Restart Forge WebUI.

🚀 Usage

  1. Launch Forge WebUI and select 🧪 LoRa x Sampler x Scheduler Grid (Forge) from the txt2img script dropdown.

  2. Choose a mode:

    • XY Grid: Select one or multiple samplers and schedulers via dropdowns. The script automatically generates all possible combinations in a comprehensive matrix format.
    • Batch Grid: Add custom sampler-scheduler pairs using dropdowns with the "Add Pair" button or manually enter pairs in the text field (format: Sampler Name, Scheduler Name per line).

  3. Configure parameters:

    • LoRA Model: Select from available LoRA models in your models/Lora directory
    • LoRA Weights: Set minimum, maximum, and step values for weight testing range
    • Trigger Word: Optional activation word with display toggle option
    • Positive/negative prompts
    • Seed (leave blank for random)
    • Steps, CFG scale, image dimensions, padding
    • Save format (WEBP/PNG)
    • Label positions for LoRA, sampler, and scheduler (Top/Bottom/Left/Right)
    • Label visibility options
    • Individual cell saving option (saves to outputs/cells/)

  4. Click "Generate" to create the grid. Generation progress for each combination and any errors are logged in the terminal. Use the "🛑 Stop Generation" button to interrupt long-running operations.

📁 Output Structure

outputs/
└── 🧪 LoRa x Sampler x Scheduler Grid (Forge)/
    ├── cells/                                  # Individual cell images
    │   ├── LoRAName_W0.50_Sampler_Scheduler_0_0.webp
    │   └── ...
    ├── xy_grid_1700000000_001.webp             # XY Grid output (WEBP format)
    ├── xy_grid_1700000000_001.png              # XY Grid output (PNG format)
    ├── batch_grid_1700000000_001.webp          # Batch Grid output (WEBP format)
    └── batch_grid_1700000000_001.png           # Batch Grid output (PNG format)

🛠 Requirements

  • Stable Diffusion Forge
  • Python 3.7+
  • Libraries: gradio, Pillow (included with Forge)

⚡ Technical Features

  • Auto-Correction: Intelligent name matching for samplers and schedulers.
  • Validation: Comprehensive configuration validation with helpful error messages.
  • Unicode Support: Proper handling of special characters and emojis in names.
  • Font Caching: Optimized performance through font caching mechanisms.
  • Memory Management: Efficient handling of large grids with auto-scaling.

⚠️ Limitations

  • Mode Restriction: Works exclusively in txt2img mode.
  • Resource Intensive: Large grids may require significant VRAM and processing time.
  • Font Dependency: Optimal text rendering requires Barlow-SemiBold.ttf availability.
  • Grid Size: Very large combinations may produce extremely large output files.

📜 License

MIT License © 2025 - Free to use with attribution.

🙌 Acknowledgments

  • Built with Gradio and Pillow libraries
  • Developed with AI assistance tools
  • Inspired by the needs of the Stable Diffusion artist community

🤝 Contributing

Feedback, bug reports, and feature suggestions are welcome. Please report issues with detailed descriptions of the problem and your configuration.

📸 Examples


🧩 Batch Grid
🔄 XY Grid

📊 Small Steps Grid

⚡ With Trigger Word
🚫 Without Trigger Word

🐋 Huge Grid