内容
<p align="center">
<img src="docs/assets/readme-hero.svg" alt="Kohya-LoRA Tool — one workspace, four training engines" width="100%" />
</p>
<h1 align="center">Kohya-LoRA Tool</h1>
<p align="center">
<a href="https://github.com/l1934332574-maker/Kohya-LoRA-Tool/releases/latest"><img alt="Latest release" src="https://img.shields.io/github/v/release/l1934332574-maker/Kohya-LoRA-Tool?style=flat-square&color=6366f1" /></a>
<img alt="Windows 10 / 11" src="https://img.shields.io/badge/Windows-10%20%2F%2011-2563eb?style=flat-square" />
<a href="LICENSE"><img alt="MIT License" src="https://img.shields.io/badge/License-MIT-059669?style=flat-square" /></a>
</p>
<p align="center">
<a href="README.zh-CN.md">简体中文</a> ·
<a href="https://github.com/l1934332574-maker/Kohya-LoRA-Tool/releases/latest">Download</a> ·
<a href="https://modelscope.cn/models/FGtiancai/Kohya-LoRA-Tool">China mirror</a> ·
<a href="CHANGELOG.md">Changelog</a> ·
<a href="https://github.com/l1934332574-maker/Kohya-LoRA-Tool/issues">Issues</a>
</p>
Train character, style and concept LoRAs on Windows with **four engines in one workspace**.
Chinese UI · NVIDIA GPUs · Selected AMD ROCm configurations
## Features
- **Unified training** — kohya, musubi-tuner, AI Toolkit and Fizgig.
- **Guided setup** — engine installation, model downloads and local model selection.
- **Dataset tools** — WD14 / PixAI tagging, caption editing and image-to-prompt tools.
- **Training and review** — presets, project management, fixed-seed samples and A/B comparisons.
- **AI assistance** — optional in-app assistant and local MCP access for external agents.
- **Concept sliders** — text-guided and image-pair LoRA training for SDXL and standard Anima; experimental.
## Supported models
| Engine | Models |
| --- | --- |
| **kohya / sd-scripts** | SD 1.5, SDXL, FLUX.1, Anima |
| **musubi-tuner** | Krea 2, FLUX.2 |
| **AI Toolkit** | Krea 2, Qwen-Image, Z-Image Base, MiniMax H3 |
| **Fizgig** | Krea 2, FLUX.2 Klein 9B, Qwen-Image-2.1, Z-Image Turbo, MiniMax H3; experimental SDXL / standard Anima |
MiniMax H3 also supports video through AI Toolkit and image / video / audio through Fizgig. Model versions, GPU requirements and available features vary by engine; see the [user guide](README_使用说明.md).
## Quick start
1. Download `Setup.exe` from [Releases](https://github.com/l1934332574-maker/Kohya-LoRA-Tool/releases/latest) or [ModelScope](https://modelscope.cn/models/FGtiancai/Kohya-LoRA-Tool). For the portable edition, extract the ZIP and run `Kohya一键工具.exe`.
2. Create a project, select a model and follow the setup guide.
3. Import images, review captions and parameters, then train.
## Resources
[User guide](README_使用说明.md) · [MCP](docs/MCP.md) · [Development](modern_ui/README.md) · [Changelog](CHANGELOG.md)
**QQ group: 602396066** · [Report an issue](https://github.com/l1934332574-maker/Kohya-LoRA-Tool/issues)
<details>
<summary>Run from source</summary>
Python 3.12 · Node.js 20.19+ or 22.12+ · WebView2
```powershell
python -m pip install customtkinter pillow -r requirements-ui.txt
npm --prefix modern_ui ci
npm --prefix modern_ui run build
python kohya_gui.py
```
</details>
## License
[MIT](LICENSE) · [Third-party notices](THIRD_PARTY_NOTICES.md)
