an attempt at training Ulquiorra Cifer from the show bleach, if you guys can find out the best settings please let me know
Description
Comments (2)
Haven't tried it yet, but always happy for more Bleach LoRas!
Most important is good quality dataset. Captioning I'm still trying to find the best tactic for, but I use https://github.com/starik222/BooruDatasetTagManager for help though. Not the autotagger, but it's useful for captioning in general. I like using the comfy flux trainer nodes rather than onetrainer or kohya proper. Also, I think it's good to get 'scenes' rather than character portraits. Hard to make characters move around when all their images is their head and shoulders..
Can't wait to try it out and maybe see more Bleach ones!
My process is usually find a bunch of high quality photos of varying poses/styles, then use the git captioner and manually edit them to be correct, then add tags after, finnally run in kohya, issue with this one may be too much variation cause of the forms, probably wouldve ran better if i just did a model for each specific form