Large training data, wich is the Best way to use it for a model?
I spent a lot of time collecteing the bests images for a model (style model) that i have always wanted to do. Now I'm with 6 folders for specific little dettails and every folder with 20-30 images, and a 141 main folder with full body view and other stuff (for a total of 350 images), and i would like to dont spend 8 hour for every test to see what is the Best way to train this model. If Anyone as some tips like if i should use lora or Lycoris, how many repeats in the train and things like that?
When you said large, I was expecting thousands! I wouldn't bother with a finetune or full checkpoint with anything under 1k images, so Lora or Lycoris methods are the way.
You said it's a style training right? Are each of the details in folders meant to be separately called, or just more refinement for the overall style?
Oh, rereading, it seems this is a character not a style?