Skip to content

Latest commit

 

History

History
100 lines (73 loc) · 3.46 KB

DEARPYGUI_UI.md

File metadata and controls

100 lines (73 loc) · 3.46 KB

Dear PyGui UI for OneTrainer

OneTrainer now includes an experimental UI implementation using Dear PyGui, which offers several advantages:

  • Improved performance with GPU acceleration
  • Modern, sleek interface design
  • Better handling of UI components like dropdowns and presets
  • Cross-platform consistency

Installation

To use the Dear PyGui UI, you need to install the additional dependencies:

# On Windows
install_dpg.bat

# On Unix-based systems
./install_dpg.sh

This will install Dear PyGui and other required dependencies for the UI.

Running the Dear PyGui UI

To start the Dear PyGui UI:

# On Windows
start-dpg-ui.bat

# On Unix-based systems
./start-dpg-ui.sh

Features

The Dear PyGui UI implementation includes all the features of the original CustomTkinter UI, but with a more modern and responsive interface:

  • Concept Management: Create, edit, and manage concepts for training
  • Model Selection: Select and configure models for training
  • LoRA Training: Configure and train LoRA models
  • Sampling: Sample models during training
  • Cloud Training: Configure and manage cloud training

Concept Management

The Concept tab allows you to create, edit, and manage concepts for training. Each concept consists of:

  • General settings: Name, path, training/validation options
  • Image augmentation: Options for augmenting images during training
  • Text augmentation: Options for augmenting text prompts during training
  • Statistics: Information about the files in the concept folder

Using the Concept Tab

  1. Add a Concept: Click "Add Concept" to create a new concept
  2. Edit a Concept: Click "Edit" on an existing concept to open the Concept Editor
  3. Clone a Concept: Click "Clone" to duplicate an existing concept
  4. Enable/Disable: Toggle the "Enabled" switch to include/exclude a concept during training

Concept Editor

The Concept Editor allows you to configure all aspects of a concept:

General Tab

  • Name: The name of the concept
  • Path: The folder containing training data
  • Enabled: Whether the concept is used in training
  • Validation concept: Use concept for validation instead of training
  • Prompt Source: Source of prompts (file per image, single file, filenames)

Image Augmentation Tab

  • Crop Jitter: Enable random cropping
  • Random Flip: Enable random horizontal flipping
  • Random Rotation: Enable random rotation with angle control
  • Brightness/Contrast/Saturation/Hue: Color adjustments
  • Resolution Override: Force a specific resolution

Text Augmentation Tab

  • Tag Shuffling: Randomize the order of tags in prompts
  • Tag Dropout: Randomly remove tags from prompts
  • Capitalization: Randomly change capitalization of tags

Statistics Tab

  • Summary: Total size, file counts
  • Image Files: Resolution stats, aspect ratios
  • Text Files: Caption stats, tag statistics

Troubleshooting

If you encounter issues with the Dear PyGui UI:

  1. Make sure you have installed the required dependencies using install_dpg.sh or install_dpg.bat
  2. Check the console output for any error messages
  3. Try running the standalone test script to check if the UI components work correctly: 
    python standalone_concept_test.py
  4. If the issue persists, report it on the Github Issues page, including the output of the debug report and steps to reproduce the issue. EOF < /dev/null