LoRA (Low-Rank Adaptation)

Low-Rank Adaptation — a fine-tuning technique that doesn't retrain the whole model. Instead, it trains a small "patch" (an adapter) and attaches it to the original model. It's like remodeling a single room instead of reconstructing the entire building, cutting costs to a tiny fraction of the original.

This is what made it possible to customize models even on a personal PC. It's especially popular in image-generation communities, where people create and share LoRAs for specific art styles or specific characters. That same popularity is why disputes over style imitation and likeness rights tend to swirl around this term.

LoRA is itself a type of fine-tuning, so asking 'which one should I use' is a bit of a false choice. Fine-tuning is the broad category, and LoRA is a cheap way of doing it. When articles today say a model was 'fine-tuned,' it usually actually means LoRA.