Skip to content

STRUCTURE.md

Latest commit

 

History

History
106 lines (84 loc) · 4.88 KB

STRUCTURE.md

File metadata and controls

106 lines (84 loc) · 4.88 KB

Project Structure

This document outlines the complete structure of the SD3.5 LoRA Training project.

stable-diffusion-3.5-text2image-lora/
├── 📄 README.md                          # Main documentation
├── 📄 LICENSE                            # Apache 2.0 license
├── 📄 requirements.txt                   # Python dependencies
├── 📄 STRUCTURE.md                       # This file
│
├── 🐍 train_text_to_image_lora_sd35.py  # Main training script
├── 🎨 inference.py                       # Image generation script
│
├── 📁 scripts/                           # Training and setup scripts
│   ├── 🔧 setup.sh                      # Environment setup script
│   ├── 🚀 train_basic.sh                # Basic training script
│   ├── 🔥 train_advanced.sh             # Advanced training script
│   └── 🎨 inference.sh                  # Easy inference script
│
├── 📁 examples/                          # Example data and templates
│   └── 📁 dataset/                      # Example dataset structure
│       ├── 📄 README.md                 # Dataset documentation
│       ├── 📄 metadata.jsonl            # Example metadata file
│       └── 📁 images/                   # Directory for training images
│           └── 📄 .gitkeep              # Placeholder for git
│
└── 📁 outputs/                          # Training outputs (created during training)
    ├── 📁 sd35-lora-basic/             # Basic training results
    ├── 📁 sd35-lora-advanced/          # Advanced training results
    └── 📁 validation_images/            # Generated validation images

📁 Directory Descriptions

Root Files

  • README.md: Complete project documentation with setup and usage instructions
  • LICENSE: Apache 2.0 open source license
  • requirements.txt: All Python package dependencies
  • train_text_to_image_lora_sd35.py: The main training script with full SD3.5 LoRA implementation
  • inference.py: Comprehensive image generation script with trained LoRA adapters

Scripts Directory (scripts/)

Contains ready-to-use bash scripts for different scenarios:

  • setup.sh: Automated environment setup (dependencies, accelerate config, directories)
  • train_basic.sh: Simple training script with sensible defaults
  • train_advanced.sh: Full-featured training with all optimizations enabled
  • inference.sh: Easy-to-use script for generating images with trained LoRA models

Examples Directory (examples/)

Provides templates and examples for users:

  • dataset/: Example dataset structure showing the required format
  • metadata.jsonl: Sample metadata file with diverse image descriptions
  • images/: Placeholder directory where users add their training images

Outputs Directory (outputs/)

Created automatically during training, contains:

  • Model checkpoints: Saved at regular intervals
  • LoRA weights: Final trained adapters
  • Validation images: Generated during training for monitoring
  • Training logs: TensorBoard logs and metrics

🚀 Quick Navigation

Want to... Go to...
Understand the project README.md
Start training immediately scripts/train_basic.sh
See example dataset format examples/dataset/
Customize training parameters train_text_to_image_lora_sd35.py
Set up environment scripts/setup.sh
Advanced training features scripts/train_advanced.sh
Generate images with LoRA scripts/inference.sh
Custom inference parameters inference.py

📋 Getting Started Workflow

  1. Setup: Run bash scripts/setup.sh
  2. Prepare Data: Add images to examples/dataset/images/
  3. Train: Run bash scripts/train_basic.sh
  4. Generate: Run bash scripts/inference.sh
  5. Monitor: Check outputs/ for results and generated_images/ for outputs
  6. Iterate: Adjust parameters and re-train as needed

🔄 File Creation During Usage

When you run the training:

outputs/
├── sd35-lora-{config}/
│   ├── pytorch_lora_weights.safetensors   # Trained LoRA weights
│   ├── adapter_config.json                # LoRA configuration
│   ├── logs/                              # TensorBoard logs
│   ├── validation_images/                 # Generated validation images
│   └── checkpoint-{step}/                 # Training checkpoints

This structure ensures easy navigation, clear organization, and smooth user experience from setup to training completion.
⚡