SimpleTuner

Why we included this project

SimpleTuner is aimed at teams that want a diffusion model trained on their own images, video, or audio rather than just a tool that generates from someone else's checkpoints. You point it at a dataset and it fine-tunes familiar open families like Stable Diffusion, SDXL, FLUX, and the newer ACE and Chroma lines into a custom checkpoint or a LoRA. The real value is that it bundles the fiddly training techniques into one place, including aspect bucketing, embedding caching, and DeepSpeed or FSDP2 sharding so a single consumer GPU can handle the job. It also adds a web UI and a job queue that supports multiple users, quotas, SSO, and worker orchestration, which makes it practical for a small team to share rather than each person running scripts alone. If you want to keep training on your own hardware instead of renting API time, this is a solid starting point.