Glossary · technique
LoRA Training
Also known as: LoRA fine-tuning, low-rank adaptation
Definition
LoRA (Low-Rank Adaptation) training is a parameter-efficient fine-tuning technique that adds small trainable matrices to a frozen base model, letting you teach an AI generator a new subject, style, or character without retraining the whole model.
In more detail
A LoRA file is typically 30–200 MB versus the multi-gigabyte size of a full fine-tune. They're ideal for consistent characters in image generation, brand-specific style transfer, or training a model to render your face. Mobbi's Mimic tool uses LoRA-style techniques to train reusable references in 10–20 minutes.
Try it on Mobbi
FAQ
How many photos do I need to train a LoRA?+
For a face: 15–30 sharp photos under varied lighting. For a style: 20–50 reference images that share the artistic look you want to teach.