# What is LoRA?

> Low-Rank Adaptation

It is a technique of making a large artificial intelligence model specialize in a particular subject by updating only a small part of it without changing the entire model.

## Overview
LoRA is a method that reduces the massive processing power required to train a massive AI model. You train only a very small layer of the model to add a new style or information to it while preserving its basic capabilities.

*Analogy: Instead of rewriting a huge library, it's like adding a small sticky note with just a few important notes on it.*

## How it works
You freeze the weights of the model and just train a small file called LoRA. By installing this file over the main model, you customize it.

## Where it is used
It is used in the process of creating personalized visuals or developing special text styles.

## Commonly confused with
Mixed with fine-tuning; While fine-tuning can cover the entire model, LoRA is much more lightweight and focused.

## Frequently asked questions
**Does using LoRA slow down the model?**
No, it generally does not cause any performance loss because it is very light.

**Can more than one LoRA be installed on a single model?**
Yes, different LoRA files can be combined for different features.


## Related terms
- [Fine-tuning](/en/dictionary/fine-tuning/)
- [AI Models](/en/dictionary/ai-models/)
- [Generative AI](/en/dictionary/generative-ai/)

## Related tools
- [LTX 2](/en/discover/ltx-2/)

---
Source: TreScout Dictionary · https://trescout.com/en/dictionary/lora/