LoRA for the character Fey from Derpixon’s Fandeltales video.
Many won’t know who this is but I’m sharing for those who do, or care. It turned out well I think. I trained it with 74 screencaps I took and edited at 19 epochs with a dim/alpha of 64/32.
Notes:
My experience has shown scale 0.5 ~ 0.8 works best, but my best results were 0.58. It gets more “cartoony” at higher scales, it looks good but the linework becomes bad; play around with it.
My prompt order shown in the sample pictures worked best for me, keep in mind the position of your <lora> matters.
To trigger the character “Fey” I reliably called the following after the loRA : “Fey_Fandeltales , green hair bow, bangs, 1girl, solo, orange eyes,blonde hair, long hair, (armor, ponytail, shoulder pads) , parted bangs, small breasts“. Add clothing descriptions to negatives to generate alternate outfits; refer to pictures.
Not intended, but this can be used to get a Derpixon style if you ignore the above bullet point and instead describe what you want or even place a different character textual inversion. Refer to the last two pictures.
I used FoxyaNSFW model, but this should work with others.
Doesn’t seem to play well with certain other LoRAs, it could be this LoRA’s fault or the other ones I was using, but textual inversions are good.
Refer to my pictures for examples and what upscaler I used, keep in mind I have used other textual inversions to help with poses and negatives.
Meant for nsfw so if you want to use this for sfw, remember to put nsfw in negative prompts etc.
Even if you don’t use it for the character, I happen to like the aesthetics regarding colour, outlines etc so it can be nice to play around with. If so: add, “green hair bow, bangs, orange eyes, blonde hair, long hair, (armor, ponytail, shoulder pads), parted bangs, small breasts ” to negatives.
Kohya LoRA trainer that I used for training, link.