Skip to content
Get packs

AI Toolkit

A toolkit for training (finetuning) image, video and audio diffusion models, with a GUI or CLI.

GitHub Download source
Stars
12.5k
Forks
1.6k
Licence
MIT
Language
Python
Last update
6 days ago
Type
App

What it is

AI Toolkit is a free, open-source suite for training diffusion models, the type of AI that generates images, video and audio. The README says it aims to support the latest models on consumer-grade hardware. You can run it through a web-based GUI or from the command line.

  • Finetune an image model such as FLUX.1-dev or SDXL on a custom set of images
  • Train a video model such as Wan 2.2 or LTX-2 on your own clips
  • Train an instruction or edit model such as Qwen-Image-Edit using control images
  • Run training from the browser UI at localhost:8675 or from the CLI

Great for

  • People who want to finetune an image model such as FLUX, SDXL or Qwen-Image on their own data
  • Users with a single consumer Nvidia GPU who want to train models locally
  • People who want a GUI for training but also the option of a CLI
  • Developers experimenting with video models such as Wan 2.2 or LTX-2

Not ideal for

  • People who only want to generate images with existing models, since this is a training tool
  • Users without an Nvidia GPU, because the manual instructions list an Nvidia GPU as a requirement and the README does not describe other hardware paths
  • Beginners with no Python or command-line experience, since the setup involves git, Python environments and PyTorch
Step-by-step guide: how to use AI ToolkitInstall, first run, examples, tips and FAQs

How to get it

  1. 1

    Clone and build from source

    You'll need: git · Python 3.10 or newer (3.12 recommended) · Nvidia GPU · Python venv

    Clone the repository

    git clone https://github.com/ostris/ai-toolkit.git

    Enter the folder

    cd ai-toolkit

    Start the AI Toolkit Manager on Linux

    chmod +x run_linux.sh

    Run the Linux launcher

    ./run_linux.sh
  2. 2

    Run it

    • Clone the repository and enter the ai-toolkit folder
    • Run the launcher script for your platform, or use python3 -m manager install
    • Start the UI with python3 -m manager launch or the launcher
    • Open http://localhost:8675 in your browser
    • Use python3 -m manager doctor if something fails
  3. 3

    Get the most out of it

    • Use python3 -m manager update to pull updates and sync dependencies
    • Check the Supported Models list to pick a model that fits your hardware and goal
    • Use the manager doctor command to diagnose setup problems
    • Use control images with Qwen-Image-2.1 if you want it to act as an editing model

Downloads

Source code (.zip)main branch · straight from GitHub

Strengths

  • Covers many recent image, video, edit and audio models in one tool, as listed in the README
  • Offers both a GUI and a CLI
  • Free and MIT licensed

Trade-offs

  • The README describes the one-click AI Toolkit Manager as experimental
  • Some model entries have caveats, such as YuE2 lacking an official audio encoder, so results may vary by model
  • The README does not state minimum VRAM for each model, so hardware needs must be checked per model

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.