What is LoRA?

How it works

LoRA (Low-Rank Adaptation) is a method for efficiently customizing AI models. Instead of retraining the entire model (which requires enormous compute resources), LoRA trains a small adapter that modifies specific parts of the model's behavior. This allows you to teach a model new concepts, styles, or subjects with just a few example images and modest compute requirements. In practice, LoRA training lets you create custom styles (train on your brand's visual language), custom subjects (train on a specific product or character), or custom techniques (train on a particular photographic style). The resulting LoRA file is small and can be shared, combined with other LoRAs, or swapped in and out. Self-hosted tools like LTX Video and Flux Schnell support LoRA training.