ostris/ai-toolkit

The ultimate training toolkit for finetuning diffusion models

What it solves

AI Toolkit is an all-in-one training suite designed to make training diffusion models accessible on consumer-grade hardware. It simplifies the process of fine-tuning image, video, and audio models, providing a comprehensive set of tools to handle the training pipeline from dataset preparation to model saving.

How it works

The toolkit provides both a Command Line Interface (CLI) and a web-based Graphical User Interface (GUI) to manage training jobs. Users can configure training runs using YAML files, where they can specify model types, network settings (such as LoRA or LoKr), and dataset paths. The system automatically handles image resizing and aspect ratio bucketing, removing the need for manual cropping.

Who it’s for

AI artists, developers, and researchers who want to fine-tune diffusion models (like FLUX, SDXL, and Wan 2.1) on their own hardware or via cloud providers like RunPod and Modal.

Highlights

  • Broad Model Support: Supports a wide array of image, instruction/edit, video, and audio diffusion models, including the FLUX family, SDXL, and Wan 2.1.
  • Flexible Training Methods: Supports LoRA and LoKr training, with the ability to target or exclude specific model layers using only_if_contains and ignore_if_contains filters.
  • User-Friendly Interface: Includes a web UI for starting, stopping, and monitoring jobs with optional authentication for secure remote access.
  • Hardware Accessibility: Optimized for consumer-grade GPUs and provides easy integration with cloud GPU providers.
  • Simplified Dataset Handling: Automatically manages image downscaling and bucketing for varying aspect ratios.