What is LoRA?
A small add-on file that teaches an image model one specific face, product or art style without retraining the whole model.
LoRA (low-rank adaptation) is a lightweight way to fine-tune. Instead of retraining a huge model, you train a small file on a handful of images; plugging it in lets the model reproduce that style or subject consistently.
LoRAs are common in open-weight image tools. Only train them on images you have the rights to — your own face, your own products, your own art.
Example
A clothing brand trains a LoRA on its new kurta collection so every AI lookbook image shows the exact print.
Related terms
Further training an existing AI model on a smaller, specific dataset so it gets better at one style or task.
An AI model whose weights are published so anyone can download, run and modify it — often for free.
Creating a picture from a written description. The core feature of ChatGPT images, Gemini, Midjourney and Flux.
