At all those web sites for using AI to create images or videos, one term appears often: "LoRa" = Low-Rank Adaptation. It turns out that these are extensions of the original model which add to the original model without that model having to be re-trained. Hereby a lot of computing time is saved.
Here is the explanation of LoRa from one of its inventors at Microsoft, Edward Hu:
In the world of generating images and videos with AI, a LoRa provides a way of extending the capabilities for behaviours or visuals of the image or video to be generated. This could refer to a specific visual style, or to particular actions that a video should show.
This is in particular useful for the generation of NSFW images and videos: here, the site https://civitai.com/ shows a huge variety of LoRas provided by users, for all kinds of situations, actions, etc., e.g. specific age of the person displayed.
