LyCORIS is an open-source library of advanced fine-tuning methods for Stable Diffusion, giving model trainers far more flexible ways to adapt image models than the standard LoRA approach. Its name spells out the idea: “LoRA beYond Conventional methods, Other Rank adaptation Implementations for Stable diffusion.” The project is maintained by KohakuBlueleaf on GitHub, has earned more than 2,500 stars, and was still actively updated in October 2026.
Plain LoRA fine-tuning works well, but it is only one point in a much larger design space of low-rank adaptation techniques. LyCORIS implements a family of alternative algorithms โ including LoHa, LoKr, and Diag-OFT โ that change how the adaptation weights are structured, which can improve capacity, convergence, or quality for specific training jobs. The library plugs into popular Stable Diffusion training workflows, so creators who already train LoRA models can experiment with stronger methods without rebuilding their whole pipeline. Everything is published as a Python package under the Apache-2.0 license.
LyCORIS is completely free and open source under the Apache-2.0 license. There is no paid tier, hosted service, or usage fee โ you only need your own GPU hardware or compute to run training jobs.
LyCORIS is for Stable Diffusion model trainers, LoRA creators, and researchers who want more expressive fine-tuning methods than the default LoRA recipe. If you publish character or style adaptations and want to squeeze more quality out of small training sets, it is one of the most established toolkits to try.