LoRA — Generative AI glossary — LiliDi
Low-Rank Adaptation — a small fine-tuning file that injects a style or character into a base model.
By LiliDi Editorial
Low Rank Adaptation — a small fine tuning file that injects a style or character into a base model. LoRA (Low Rank Adaptation) is a parameter efficient fine tuning technique that produces a small (often <100 MB) adapter file. Loaded on top of a base diffusion model, it injects a specific style, character or product without retraining the full model. Common for anime styles, brand mascots and consistent talent.