# LoRA (Low-Rank Adaptation) LoRA adds a lightweight adapter to a base model, enabling new styles or subjects without full retraining. Why it’s popular: - Small file size - Trains on consumer GPUs - Mixable with other LoRAs