SimpleTuner
A self-hosted diffusion model trainer with a web UI, LoRA and full fine-tuning, NVIDIA, AMD and Apple Silicon support, and an AGPL-3.0 license.
SimpleTuner is an open-source toolkit for fine-tuning image, video and audio generation models on your own hardware or GPU servers. It's for creators and researchers adapting models to their datasets, and teams sharing training infrastructure. A web dashboard manages training jobs.
It supports LoRA, LyCORIS and full model training across families including Flux, Stable Diffusion XL, Qwen Image, Wan Video and ACE-Step. Reference-based training covers image editing and image-to-video tasks. You can compare adapters through validation renders and exchange LoRAs in Diffusers or ComfyUI formats.
Dataset handling includes mixed image sizes and aspect ratios, disk caching to reduce repeated processing, and CaptionFlow caption generation on local GPUs. For larger jobs, multi-GPU and multi-node training distribute the workload. DeepSpeed and FSDP2 help reduce per-GPU memory demands through offloading and sharding.
Hardware support includes NVIDIA and AMD GPUs, plus Apple Silicon for LoRA training. Many models can train with memory optimizations on GPUs with 16GB or 24GB of memory; full training of larger models needs more. Docker deployment is available.
Teams can dispatch jobs to persistent or cloud GPU workers, with priority queues, access controls, SSO and audit logs. The toolkit uses the AGPL-3.0 license. It sends no data to third parties by default; external reporting, model uploads and webhooks are opt-in. Optional S3 storage connections include Cloudflare R2 and Wasabi.
Model checkpoint licenses vary; check the upstream terms separately from the toolkit license.
