LoRA (Low-Rank Adaptation)

Friendly Description: LoRA is a clever shortcut for customizing a big AI model without retraining the whole thing. Instead of changing every dial inside the model, LoRA adds a few small adjustable add-ons that nudge the model's behavior in the direction you want. It's like adding a custom lens to a camera instead of building a whole new camera; you get the new look much more cheaply and quickly.

Example: An artist might use LoRA to teach a giant image generator their personal painting style. Instead of training a brand new model, they fine-tune a tiny LoRA add-on, just a few megabytes, that, when plugged into the base model, makes everything come out in their unique look.