Easy

Easy is a command-line tool that simplifies the workflow for training and managing LoRA models with SimpleTuner. It provides a streamlined interface for configuring, training, and managing machine learning models, with a focus on diffusion models and integration with ComfyUI.

Note: This repository is independent from the comfy-download repository. If you need comfy-download functionality, please install it separately.

Features

• Simplified Configuration: Easy setup of training configurations with interactive prompts
• Dataset Management: Tools for organizing and preparing datasets
• Training Management: Easily initiate and monitor training jobs
• LoRA Model Management: Tools for moving trained models to ComfyUI
• Grid Visualization: Create validation and dataset grids to compare results
• Dropbox Integration: Sync models and configs with Dropbox for backup and sharing

Installation

Prerequisites

• SimpleTuner installed at /workspace/SimpleTuner
• ComfyUI installed at /workspace/ComfyUI
• Python 3.8 or higher

Dependencies

pip install -r requirements.txt

Required packages:

• rich
• pillow
• safetensors
• numpy

Setup

1. Clone this repository:

git clone https://github.com/yourusername/easy.git /workspace/easy

2. Run the setup script:

cd /workspace/easy
bash setup.sh

Project Structure

/workspace/easy/
├── easy.py                # Main script entry point
├── easy.bat               # Windows batch wrapper
├── easy.sh                # Linux/macOS shell wrapper
├── easy.json              # Main configuration file
├── classes/               # Core functionality classes
│   ├── config.py          # Training configuration management
│   ├── multidatabackend.py # Dataset configuration management
│   ├── userpromptlibrary.py # Prompt template management
│   ├── response.py        # UI response handling
│   ├── lora_mover.py      # LoRA model management tool
│   ├── lora_sync.py       # Dropbox sync tool
│   ├── validation_grid.py # Validation grid generation
│   ├── dataset_grid.py    # Dataset grid generation
│   └── download_configs.py # Config downloader tool
├── names/                 # Naming preset templates
├── prompts/               # Prompt templates
└── scenario/              # Training scenario templates

Usage

Easy provides a command-line interface with several commands for managing the model training workflow.

Basic Commands

# Initialize a new training configuration
easy init <instance_prompt> <version> <dataset> <scenario> <naming_preset> <id_base> <resolutions> <prompt_file>

# Edit an existing configuration
easy edit <partial config folder name> <type(config/backend)>

# Reinitialize a configuration with new parameters
easy reinit <partial config folder name> <instance_prompt> <version> <dataset> <naming_preset> <id_base> <resolutions>

# List available configurations or datasets
easy list <config/datasets> [group]

# Start training with a specific configuration
easy train <n>

# Show help information
easy help

Advanced Commands

# LoRA Mover - Process and move trained models to ComfyUI
easy lm

# LoRA Sync - Sync models with Dropbox
easy ls

# Download configurations from Dropbox (supports custom model mappings)
easy dc

# Create validation grid for model checkpoints
easy vg

# Create dataset grid to visualize training data
easy dg

# Run post-processing tools (lm, dc, vg, dg)
easy pp

# Train and run post-processing tools
easy tpp

Configuration

Main Configuration (easy.json)

The easy.json file defines the paths to various components:

{
    "simple_tuner_path": "/workspace/SimpleTuner",
    "config_folder_path": "/workspace/SimpleTuner/config",
    "scenario_folder_path": "/workspace/easy/scenario",
    "dataset_folder_path": "/workspace/SimpleTuner/datasets",
    "prompt_folder_path": "/workspace/easy/prompts",
    "names_folder_path": "/workspace/easy/names",
    "output_folder_path": "/workspace/SimpleTuner/output",
    "comfy_models_folder_path": "/workspace/ComfyUI/models"
}

Model Mappings (optional)

You can create a settings/model_mappings.json file to customize how config folders are mapped to Dropbox folders when using the easy dc command. This file is not tracked by git for privacy.

{
  "mappings": {
    "001-model1-folder": [
      "model1",
      "model1special",
      "model1new"
    ],
    "002-model2-folder": [
      "model2variant1",
      "model2variant2"
    ]
  },
  "defaults": {
    "model1": "001-model1-folder",
    "model2": "002-model2-folder"
  }
}

How Model Mappings Work

1. Config Name Extraction:

   • For a config like "phoebegrey_005_constant_1e_4_1_32", the script extracts the prefix before the first underscore ("phoebegrey")

2. Mapping Lookup Process:

   • First, checks the "mappings" section for a direct match
   • If not found, falls back to the "defaults" section to find a matching prefix
   • If a match is found in either section, uses the corresponding Dropbox folder name

3. Dropbox Folder Naming:

   • Dropbox folder names (like "001-model1-folder") can be any format you prefer
   • You can use descriptive names like "028-phoebemilo-fashionAW25-brand"
   • The script does not parse or analyze the folder name structure

4. Key Benefits:

   • Map multiple config variants to the same Dropbox folder
   • Use descriptive Dropbox folder names without affecting matching
   • Add new configs to existing folders without explicit mapping
   • Easily handle folder renaming in Dropbox

5. Use Cases:

   • Group multiple training variants under a single Dropbox folder
   • Organize by collection/season with descriptive folder names
   • Use defaults as a catch-all for configs sharing common prefixes
   • Create new Dropbox folders simply by adding them to mappings

Training Configuration

Training configurations are created interactively and stored as JSON files. Key options include:

• Instance prompt
• Optimizer settings
• Learning rate and scheduler
• Batch size
• Training steps
• Checkpoint frequency

Prompt Templates

Prompt templates are defined in JSON files in the prompts/ directory and are used to generate formatted prompts for model training. Example:

{
    "edgy_urban_fashion-{--instance_prompt}": "A bold and contemporary fashion editorial...",
    "resort_beachwear-{--instance_prompt}": "A luxurious resort fashion editorial...",
    "minimalist_high_fashion-{--instance_prompt}": "An avant-garde high fashion editorial...",
    "simple-{--instance_prompt}": "a midshotportrait of {--instance_prompt} in a simple setting...",
    "closeup-{--instance_prompt}": "a telephoto closeup portrait of {--instance_prompt}..."
}

Workflow Example

A typical workflow using Easy would be:

1. Prepare your dataset in /workspace/SimpleTuner/datasets/your_dataset

2. Initialize a new configuration:

   easy init character_name 01 your_dataset default default lora "512,768,1024" default

3. Start training:

   easy train character_name

4. When training completes, move the models to ComfyUI:

   easy lm

5. Create validation grids to compare checkpoints:

   easy vg

6. Sync your models to Dropbox for backup:

   easy ls

7. Download and backup configurations:

   easy dc

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

[Add your license information here]