/vt/ - Virtual Youtubers

>>41812783
I'm not a pro lora trainer or anything but I tried making a shion lora on multiple outfits at once by splitting them into different folders, each with a different trigger word at the start of all their tag files - https://litter.catbox.moe/7gg7ki.png ("other" is mostly nude but also just has a few pics of things that aren't any of her actual 3 outfits)

The trigger words essentially do nothing. The lora can generate pics of all the outfits, sometimes, but it takes a little bit of outfit description in the prompt to help push it toward the correct one, and even then it sometimes blends them in strange ways, like how it puts her hat or little cape thing into other outfits often. This comparison pic was generated on a model that doesn't have any hll mixed into it so basically all the character concept is coming from the lora.

Overall I feel like I would have gotten better results just making a different lora for each outfit. It's not like she can wear two different outfits at the same time anyway so just switching loras depending on what you are trying to prompt is not that big a deal?