Repolode
Submit
ostris

ostris/ai-toolkit

An all-in-one training suite for finetuning diffusion image, video and audio models on consumer GPUs, via a web UI or CLI.

View on GitHub
Stars
12,378
Forks
1,587
Open issues
92
Last commit
2 days ago

What is it?

AI Toolkit is a training suite for finetuning diffusion models. It covers image, instruction-based editing, video and audio models, including families such as FLUX, Qwen-Image, SDXL, Wan 2.x and LTX-2. The project aims to run these models on consumer-grade hardware.

Training can be driven from a web UI that starts, stops and monitors jobs, or from the command line using YAML config files passed to run.py. Training resumes from the last checkpoint after interruption. Datasets are folders of images with matching caption .txt files, and images are resized and bucketed automatically so they do not need manual cropping.

The project supports LoRA and LoKr networks, with options to target specific layers through include and exclude filters. It can be installed through a built-in manager that sets up the PyTorch environment, or manually with Python 3.10+ and an Nvidia GPU.

Key features

  • 01Trains image, edit, video and audio diffusion models from one toolkit.
  • 02Provides a web UI for starting, stopping and monitoring training jobs.
  • 03Supports command-line training from YAML config files.
  • 04Supports LoRA and LoKr networks with per-layer include and exclude filters.
  • 05Resumes training from the last checkpoint after interruption.
  • 06Automatically resizes and buckets images with varying aspect ratios.

Use cases

  • Training a custom FLUX or SDXL LoRA style or character on a single consumer GPU.
  • Finetuning a Wan or LTX video model on a personal dataset.
  • Running training jobs on a remote server and monitoring them through the web UI.

Quick start

  1. 1

    Clone the repository

    git clone https://github.com/ostris/ai-toolkit.git
  2. 2

    Enter the project folder

    cd ai-toolkit
  3. 3

    Start the AI Toolkit Manager on Linux (it sets up the environment and launches the UI at http://localhost:8675)

    ./run_linux.sh

Things to know

  • The one-click manager is marked experimental by the author.
  • Manual install requires an Nvidia GPU and enough VRAM for the model you train.
  • Some supported models, such as FLUX.1-dev, require requesting access on Hugging Face first.

Topics

#diffusion-models#lora#fine-tuning#stable-diffusion#flux#video-generation#pytorch

FAQ

What is ai-toolkit?
An all-in-one training suite for finetuning diffusion image, video and audio models on consumer GPUs, via a web UI or CLI. AI Toolkit is a training suite for finetuning diffusion models. It covers image, instruction-based editing, video and audio models, including families such as FLUX, Qwen-Image, SDXL, Wan 2.x and LTX-2. The project aims to run these models on consumer-grade hardware.
Is ai-toolkit free to use? What license does it have?
ai-toolkit is open source under the MIT license. Check the license terms before using it in commercial products.
Is ai-toolkit still maintained?
Actively maintained. The last commit to ai-toolkit was on Oct 19, 2026.
What can I use ai-toolkit for?
Training a custom FLUX or SDXL LoRA style or character on a single consumer GPU. Finetuning a Wan or LTX video model on a personal dataset. Running training jobs on a remote server and monitoring them through the web UI.
How do I install ai-toolkit?
Clone the repository: git clone https://github.com/ostris/ai-toolkit.git Enter the project folder: cd ai-toolkit Start the AI Toolkit Manager on Linux (it sets up the environment and launches the UI at http://localhost:8675): ./run_linux.sh
Where can I get ai-toolkit?
The source code and installation instructions are on GitHub: https://github.com/ostris/ai-toolkit

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects