I have been working on this model on and off for some months now, I just haven't gotten it perfect yet, and will continue to work on it. Sharing now in hopes of gathering constructive feedback from people better at training realistic LORAs than I am.

Dataset built using screenshots from The Office Seasons 1-4