Mirin Chikuwa x Ratatatat74 style mix
Description
FAQ
Comments (6)
When creating these types of artist mixes, do you simply combine both artists' datasets? Or do you do something else? I tried to do this in the past, but the lora tended to "choose" one style no a mix of them.
Maybe you just merge 2 loras?
equal count of pics of each style+gradient at settings. And pics must be with same tags +-
Ну и несколько экспериментов, что лучше работает. Сейчас у меня проблема тренировать lineart стили - у анимы как будто нет претренненого понимания и тренировки со скрипом идут
@Xegakup Sorry, what do you mean same tags? same number of tags?
@sovietlat no, same pics scenery at all
as 1girl standing white background etc
Generally speaking, there should be about the same number of important tags as 1girl, 1boy etc for each style
if there is an offset, then one style will be triggered by such tags, but the second will not.
@Xegakup I see that was probably my problem then. Thank you