This stands for Low Rank adaptation. In AI terms, this is a small add-on to the model that makes it do specific things. It is usually used in the context of stable diffusion. A large model takes very long to train with significant resources. However, sometimes, you need specific details – for example you could have a model that needs to do better image generation of the hands, in which case you train a small subset with hands and that is the LoRa that will be used specifically for hands. Remember, LoRa does not do the full model but just freezes weights and parameters. Then it adds matrices which are applied to new inputs to get specific results in the context that is needed.
LoRa is used sometimes to create images in a specific style.
In the picture above that image was generated with an image generator in the drawing style. Whereas in the fantasy or other styles it looks very different with the same prompt.
