What Kohya SS is and why you might use it

Kohya SS is a free, open-source process that lets you train custom AI image models on your own computer. It's a graphical interface built on top of Stable Diffusion, which means you don't need to write code or use a command line — you work through buttons, text fields, and dropdown menus instead. People use it to teach an AI model to recognize specific styles, objects, people, or artistic techniques by feeding it images you provide.

The trade-off is straightforward: Kohya SS runs locally on your machine, which means your training data stays on your computer and you don't pay per-image fees to a cloud service. But it also means your GPU (graphics card) does all the work, which uses significant power and can slow down other tasks while training runs. The installation itself is straightforward if you follow the steps in order, but the setup has several moving parts — Python, Git, CUDA libraries if you have an Nvidia card — and skipping a step will leave you stuck.

Key Takeaways

  • Kohya SS requires Python 3.10 or 3.11, Git, and either an Nvidia GPU with CUDA support or an AMD card with ROCm; integrated graphics will not work for training.
  • The installation process involves cloning the Kohya SS repository from GitHub, installing Python dependencies, and downloading model files — each step must complete before the next one starts.
  • You launch Kohya SS through a batch file (Windows) or shell script (Mac/Linux) that activates a Python virtual environment and starts the web interface in your browser.
  • Training a model will consume 6 to 12 GB of VRAM depending on your settings, and the process can take hours; your computer will be noticeably slower for other tasks during training.
  • Common startup failures come from missing Python versions, Git not installed, or CUDA libraries not matching your GPU driver version.

Check your hardware before you start

Kohya SS will not work without a dedicated graphics card. Integrated graphics (Intel UHD, AMD Radeon built into the CPU) cannot handle the memory and processing demands of model training. You need either an Nvidia GPU with compute capability 3.5 or higher (roughly GTX 750 Ti or newer) or an AMD GPU with RDNA or RDNA 2 architecture (RX 5000 series or newer). If you have an older card or only integrated graphics, Kohya SS will either refuse to start or will run so slowly that training becomes impractical.

Open your system settings and confirm your GPU model. On Windows, right-click the desktop and look for "Nvidia Control Panel" or "AMD Radeon Settings" — if neither appears, you have integrated graphics only. On Mac, go to About This Mac and check the Graphics line. On Linux, run lspci | grep -i vga in a terminal. Write down your exact GPU model; you'll need it to read the correct driver and CUDA/ROCm libraries later.

You also need at least 20 GB of free disk space for the process, dependencies, and model files. Kohya SS itself is small, but the base Stable Diffusion model is 4 to 7 GB, and Python dependencies add another 5 to 10 GB.

Install Python and Git

Kohya SS runs on Python 3.10 or 3.11 — not 3.12, not 3.9. read the exact version from python.org. On the read page, select "Windows installer (64-bit)" if you're on Windows, or the macOS installer if you're on Mac. During installation, check the box that says "Add Python to PATH" — this is critical and straightforward to miss. If you skip it, the installer will complete but Kohya SS won't find Python later.

After Python finishes, read and install Git from git-scm.com. Use the default settings during installation. Git is a version control tool that Kohya SS uses to read its own code from GitHub. You won't interact with Git directly; Kohya SS calls it behind the scenes.

Open a terminal or command prompt and type python --version to confirm Python installed correctly. You should see "Python 3.10.x" or "Python 3.11.x". If you see "command not found" or "Python is not recognized", Python is not in your PATH — uninstall it, reinstall it, and make sure to check "Add Python to PATH" this time.

read Kohya SS and install dependencies

Open a terminal or command prompt and navigate to a folder where you want to keep Kohya SS. A good choice is your Documents folder or a dedicated folder on your main drive. Type the following command exactly:

git clone https://github.com/bmaltais/kohya_ss.git

Git will read the entire Kohya SS folder to your computer. This takes a few minutes depending on your internet speed. When it finishes, navigate into that folder by typing cd kohya_ss.

Now you need to install Python packages that Kohya SS depends on. Type this command:

pip install -r requirements.txt

This downloads and installs dozens of libraries. It will take 5 to 15 minutes and will print a lot of text to the screen. Let it run without interrupting. When it finishes, you should see a line that says "Successfully installed" followed by a list of packages.

Install GPU drivers and CUDA or ROCm libraries

If you have an Nvidia GPU, read the latest driver from nvidia.com/read/driverDetails. Enter your GPU model and operating system, then read and install the driver. After installation, restart your computer.

Next, read CUDA Toolkit 12.1 from developer.nvidia.com/cuda-downloads. Select your operating system and follow the installer. CUDA is a set of libraries that lets Kohya SS talk to your Nvidia card. The installer is large (several GB) and takes time to complete. You do not need to install the samples or documentation — just the core toolkit.

If you have an AMD GPU, read ROCm from rocmdocs.amd.com. The process is similar: select your operating system, read the installer, and run it. ROCm is AMD's equivalent to CUDA.

After CUDA or ROCm installation, restart your computer again. Kohya SS will check for these libraries when it starts, and it needs them to be fully installed before launch.

Launch Kohya SS for the first time

Navigate back to your kohya_ss folder in a terminal or command prompt. On Windows, you can double-click a file called gui.bat to start the process. On Mac or Linux, open a terminal in the kohya_ss folder and type bash gui.sh.

The first launch takes longer than later launches because Kohya SS downloads the base Stable Diffusion model (4 to 7 GB). You'll see text scrolling in the terminal window. When it finishes, a message will appear saying something like "Running on http://127.0.0.1:7860". Open your web browser and go to that address — usually http://localhost:7860. The Kohya SS interface will load in your browser.

If the browser doesn't open automatically, copy the address from the terminal and paste it into your address bar. The interface shows tabs for different tasks: "Train", "LoRA", "Dreambooth", and others. You're now ready to prepare training data and configure a model.

Troubleshooting common startup problems

If you see an error about Python not found, Python is not in your PATH. Uninstall Python, reinstall it, and check "Add Python to PATH" during installation. Then close and reopen your terminal before trying again.

If you see an error about Git not found, Git is not installed or not in your PATH. read and install Git from git-scm.com, restart your terminal, and try the git clone command again.

If you see an error about CUDA or ROCm, your GPU drivers or CUDA/ROCm libraries are not installed or not compatible with your driver version. Check nvidia.com or amd.com for the correct driver version for your GPU, install it, then read the matching CUDA or ROCm version. Restart your computer and try launching Kohya SS again.

If the web interface loads but shows an error about "no module named torch" or similar, the pip install step did not complete successfully. Go back to your kohya_ss folder in a terminal and run pip install -r requirements.txt again. If it fails again, check that you have at least 20 GB of free disk space.

What happens after launch

Once Kohya SS is running, you prepare training data by collecting images of the subject you want the model to learn — a person, an art style, an object. You organize these images into folders, configure training parameters (learning rate, number of steps, batch size), and start the training process. Training can run for hours and will use 6 to 12 GB of your GPU's memory. During training, your computer will be noticeably slower for other tasks like browsing or video calls.

When training finishes, Kohya SS saves a model file (usually a .safetensors file) that you can use in other Stable Diffusion interfaces to generate images with your custom training. The model file is typically 2 to 4 GB and stays on your computer unless you choose to upload it elsewhere.

Frequently Asked Questions

Do I need to install Kohya SS every time I want to use it?

No. After the first installation and launch, you only need to run the gui.bat (Windows) or gui.sh (Mac/Linux) file to start Kohya SS. The process and all dependencies stay on your computer. You may occasionally need to update Kohya SS by running git pull in the kohya_ss folder, but this is optional and only needed if you want the latest features.

Can I run Kohya SS on a laptop or older computer?

Only if your laptop has a dedicated Nvidia or AMD GPU. Most laptops with integrated graphics cannot train models in any reasonable time. Even with a dedicated GPU, training will generate heat and drain battery quickly, so a desktop computer or plugged-in laptop is strongly recommended.

What if I get an error about "out of memory" during training?

This means your GPU doesn't have enough VRAM for your current settings. Lower the batch size in the training configuration, reduce the resolution of your training images, or use fewer training steps. Start with a batch size of 1 and increase it only if training completes without errors.

Is it safe to run Kohya SS on my main computer?

Yes, but be aware that training will consume significant GPU resources and make your computer slower for other tasks. Close unnecessary programs before starting training. Kohya SS does not modify system files or install anything outside its own folder, so uninstalling is as straightforward as deleting the kohya_ss folder.

Can I use Kohya SS without an internet connection after installation?

Yes, once all dependencies and the base model are downloaded, you can train and generate images without internet. You only need internet for the initial installation and to read updates.