OneTrainer describes itself as a one-stop solution for all your diffusion training needs, and its feature list backs that up. The open-source desktop application has about 3,247 stars on GitHub under the AGPL-3.0 license and is actively developed in 2026. Instead of juggling separate scripts for every model family, you train almost everything from one graphical interface.
The supported model range is unusually broad: FLUX.1 and Flux.2 variants, Stable Diffusion 1.5 through 3.5, SDXL, Qwen Image, Z-Image, Chroma, PixArt, Sana, Ernie Image, and even video models such as Hunyuan Video, in both diffusers and checkpoint formats. Training methods include full fine-tuning, LoRA, and embedding training. Practical touches fill out the tool: masked training so the model focuses on the right parts of samples, automatic backups that capture everything needed to resume seamlessly, image augmentation, TensorBoard integration for tracking progress, multiple prompts per image, EMA weights with CPU offload to save VRAM, aspect ratio bucketing, multi-resolution training, and noise scheduler rescaling based on published research. For creators who train models regularly, it replaces a folder of fragile command lines with one reproducible workflow. The interface is built for long, iterative training runs rather than quick demos, organizing concepts the way experienced trainers think about a job: dataset, model, training method, and monitoring, all in one place and saved as reusable configurations.
OneTrainer is free and open source under the AGPL-3.0 license.
OneTrainer is for AI artists, model trainers, and small studios who regularly fine-tune image or video diffusion models and want one reliable, full-featured training application.