← Glossary Term
LoRA
A lightweight way to teach an existing AI model a new style or subject without retraining the whole thing.
LoRA, short for Low-Rank Adaptation, is a small add-on file that nudges a big AI model in a new direction. Instead of retraining the entire model, which takes huge amounts of computing power, you train only a thin layer of extra settings on top. Think of it as putting a lens on a camera rather than building a new camera.
In image generation, people use LoRAs to teach a model a particular art style, a character or a product from just a handful of pictures. The files are small, easy to share and can be switched on or off, which is why they are so popular with hobbyists running models on their own computers.
Mentioned in