Diffusers Tuner
A lightweight, configuration-driven fine-tuning framework for Hugging Face Diffusers pipelines, with a simplified LoRA implementation and native distributed training.
Diffusers Tuner aims to remove the overhead of rewriting diffusion-model training pipelines from scratch and of managing complex LoRA injections and configuration by hand. It provides a data preparation step, a tuning interface, and a configuration system built on Hydra for hierarchical, config-first workflows, along with native distributed data-parallel support.
Its LoRA implementation is deliberately simplified, exposing mainly the rank and target modules rather than the complexity of larger LoRA libraries. A dataset schema adapts to different dataset formats, and bucketed, batched dataset training supports larger batch sizes across different image resolutions.
It ships specialized support for Qwen-Image and FLUX.1, including the FLUX.1-Fill-dev pipeline and Qwen-Image-Edit-2511. It targets Python 3.12, PyTorch 2.10 or newer, and Diffusers 0.37.0 or newer, and installs with uv or pip; the project describes itself as still in progress.