Kohya_ss
Gradio GUI for training LoRAs and fine-tunes with Kohya's diffusion scripts.
About
Kohya_ss is the standard graphical front end for Kohya's sd-scripts, the training toolkit behind most community LoRAs. Instead of hand-writing long training commands, users configure datasets, learning rates, and schedulers in a Gradio web UI that generates the CLI commands for them. It covers LoRA, LoHa, LoKr, DreamBooth, full fine-tuning, textual inversion, and concept-erasure training.
The project tracks the ecosystem's model releases, supporting Stable Diffusion 1.5/2.x, SDXL, SD3, Flux.1, Lumina Image 2.0, and HunyuanImage-2.1. It remains the default recommendation in tutorials for training custom styles and characters, with Linux support actively maintained and Windows as the primary platform.
What it can do
- LoRA, LoHa, LoKr, and DreamBooth training
- Fine-tuning and textual inversion
- Support for SD, SDXL, SD3, and Flux.1
- GUI that generates training CLI commands
- Concept erasure (LECO) training
Setup
Clone bmaltais/kohya_ss from GitHub and run the setup script for your OS.
Prerequisites
Python 3.10-3.11, a CUDA-capable NVIDIA GPU, and a captioned training dataset.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 97; Trending score 72/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.