@sheerbabes742 per your request for feedback.

2 big issues, the first of which is your tiny dataset (5 images). With that small a dataset, there are going to be additional similarities between the pictures that are learned as well, that may sometimes take precedence over what you're trying for. Case in point, this appears to be using standard checkpoint definition "daisy dukes" most of the time, and very rarely overrides that with your skimpier "daisy dukes shorts". It may think that denim is 'daisy dukes shorts', rather than daisy dukes being 'daisy dukes shorts'. That leads into the second issue, your trigger. Don't use a term that the checkpoint already knows as your trigger (or part of your trigger), it confuses the AI, and you'll need a high weight to reliably override the checkpoint's learning (it's part of why loras often have word mashups, author names, etc as their triggers)