/vt/ - Virtual Youtubers

>>75310704
What do you mean?
Are you asking what you should do for tagging in the dataset for a LoRA?
Some suggest tagging everything describing the character is a more fluid training method, but I always tag the character and remove defining characteristics from the tagging. If the character always has defining features (e.g. animal ears/hair color/etc), you can remove it from the auto-generated captions. Adding it in the prompts will sometimes help, but I always remove it from the dataset captions.
For specific outfits, you can replace tags describing that outfit and concatenate CLIP-distinguished Tokens (mintjacket) and prompt additional words to help clean up gens + edit/inpaint to fix inaccuracies in design.

If this isn't what you meant, maybe clarify the question?

>>75251507
grazi
I don't come to /vt/ frequently, but it's nice to cross-pollinate resources sometimes. I like /vt/'s passion.