AI Toolkit (ostris)

AI Toolkit (ostris) is an MIT-licensed training suite for people who want to fine-tune image and video models on their own hardware or a self-hosted server. It targets consumer NVIDIA GPUs and runs on Linux and Windows, including ARM64 Linux systems such as DGX Spark. An experimental installer also supports Apple Silicon Macs. GPU memory needs depend on the model and training task.

Its model support covers FLUX, SDXL and Stable Diffusion 1.5, alongside Qwen-Image and image-editing models such as FLUX.1-Kontext-dev. Video training includes Wan, LTX and MiniMax-H3, with support for reference-image-to-video training. It also supports audio models such as ACE-Step and the multimodal Qwen2.5-Omni model.

You can train through a browser interface or the command line. The web UI lets you start, stop and monitor jobs, and training continues when the interface isn't running. Token authentication can restrict access to a server's UI. Checkpoints let interrupted training resume rather than start over.

For fine-tuning, it supports LoRA and LoKr adapters, including control over which model layers train. Its image loader handles different aspect ratios and resizes images for batching, so datasets don't need uniform crops. Training outputs include checkpoints and sample images.

Local training writes outputs to your machine. GPU rental through Ostris Cloud, RunPod or Modal moves training to cloud hardware; Modal stores models and samples in cloud storage. Access to FLUX.1-dev through Hugging Face requires an account, an access token and model approval.