This is an unnamed character that exists for approximately 30 seconds in one scene of one episode of The New Batman Adventures (2x10 "Beware the Creeper"). As such, it's trained on 8 cropped screencaps of that scene.
0.6 - 0.8 weight plays nicest
helper tags: black bob haircut, bangs, hair band, makeup
outfit: teal sweater, black pants
put "clothing store" or some permutation thereof in negative to reduce the occurrence of mannequin heads and other retail paraphernalia, since, again, there's less than a minute of footage to work off of
Description
FAQ
Comments (3)
And heres me thinking my 15 images of a character might be too litttle... huh.. interesting. Wasnt aware that a Lora of a stylized character could be trained by just 8 images. How many epochs etc did you run on those 8 images? Did you stealth add anymore images like crop heads off other characters to give the training more datasets?
Also what base model did you run the training on for BATS character? I tried a few a few and they don't mesh well with the base model styles.
I just use the Google Colab trainer because my computer is too old to crunch numbers and I am too dumb to actually figure out the real nuts and bolts of all this. I ran it through AnyLora for 2K steps, which turned out to be 50 epochs.
Thriftie, one of /co/'s earliest FotMs. Nice