ComfyUI + AI-Toolkit integrated environment for LoRA training and AI image/video generation workflows.
| Component | Description |
|---|---|
| Training Guides | Step-by-step guides for AI Toolkit LoRA training, FLUX.2 workflows, HunyuanImage-3.0 workflows (Chinese) |
| Training Scripts | BAT/Python scripts to launch training, Web UI, and shell environments |
| Workflow Examples | Ready-to-use ComfyUI workflow JSONs (FLUX, LTX, WAN, Seedance, etc.) |
| Video Generation Tools | LTX 2.3 i2v batch processing, Seedance i2v/t2v scripts with web interface |
| AI Toolkit Web UI | Custom web interface for managing LoRA training projects |
| Example Training Projects | FLUX LoRA project config with dataset and trained model output |
This repo provides the custom scripts, configs, and documentation layer. You need to set up the following separately:
- ComfyUI Portable - Download from GitHub
- AI-Toolkit - Clone ostris/ai-toolkit into the
ai-toolkit/directory - NVIDIA GPU with sufficient VRAM (16GB+ for LoRA training, 32GB+ recommended)
# Option 1: Double-click the BAT script
run_ai_toolkit_train.bat
# Option 2: Use the Web UI
run_ai_toolkit_web.bat
# Open http://localhost:7860# Standard launch
run_nvidia_gpu.bat
# With CORS support (for external API access)
run_nvidia_gpu_cors.bat# LTX 2.3 Image-to-Video batch processing
# Open ltx23_i2v_batch_web.html in browser
# Seedance Image-to-Video
# Open seedance_i2v_batch_web.html in browser├── AI_TOOLKIT_训练指南.md # AI Toolkit training guide
├── FLUX2_DEV_工作流程指南.md # FLUX.2 [dev] workflow guide
├── HunyuanImage3_工作流程指南.md # HunyuanImage-3.0 workflow guide
├── ai_toolkit_web.py / .html # Training management Web UI
├── ai_toolkit_projects/ # Training project configs & datasets
│ └── FLUX/
│ ├── config.yaml # Training configuration
│ └── dataset/ # Training images + captions
├── AI_Training_data/ # Training data & outputs
│ ├── Spartan/ # Source training images
│ └── output/ # Trained LoRA models (.safetensors)
├── workflows_example/ # ComfyUI workflow JSON files
├── run_*.bat # Launch scripts
├── ltx23_i2v_batch_*.py/.html # LTX video generation tools
├── seedance_*.py/.html # Seedance video generation tools
└── ComfyUI/ # (not included - install separately)
| Model | Type | Guide |
|---|---|---|
| FLUX.2 [dev] | Image Generation | FLUX2_DEV_工作流程指南.md |
| HunyuanImage-3.0 | Image Generation + Editing | HunyuanImage3_工作流程指南.md |
| Qwen Image / Image Edit | Image Generation + Editing | AI_TOOLKIT_训练指南.md |
| WAN 2.1/2.2 14B | Video Generation | AI_TOOLKIT_训练指南.md |
| LTX 2.3 | Video Generation (i2v) | Included batch scripts |
| Seedance | Video Generation (i2v/t2v) | Included batch scripts |
| Task | Minimum VRAM | Recommended |
|---|---|---|
| LoRA Training (FLUX/Qwen) | 16 GB | 24-32 GB |
| LoRA Training (WAN 14B) | 24 GB | 32 GB |
| FLUX.2 Inference | 32 GB | 48+ GB |
| HunyuanImage-3.0 Inference | 48 GB | 96+ GB |
Training scripts and documentation are provided as-is for educational purposes. Model weights are subject to their respective licenses (FLUX, Qwen, WAN, etc.).