GLOSSARY
LoRA (Low-Rank Adaptation)
A lightweight way to teach a base model a new style, character or skill — a small add-on file trained in hours instead of retraining the whole model.
LoRA fine-tunes a model by training only a tiny set of extra weight matrices that attach to the original model, rather than updating billions of parameters. The result is a file of a few megabytes (versus gigabytes for a full model) that dramatically changes output — a particular art style, a consistent character, a product look.
The open image-model ecosystem runs on LoRAs: platforms like Tensor.Art host thousands of community-made ones you can combine with base models on the fly. For teams, LoRA is the cheapest way to keep a house style consistent across generated assets without retraining anything.