Thanks to pyramidMatriarch for providing their dataset, their version of the model linked below.
Main is the real version, Wojack was made as a joke.
Main version produces an anime girl that looks like the doomer girl meme, Wojack version produces a wojack.
Both versions work at 0.8 and share trigger words.
Note: example images using style loras have that lora's name in [] brackets.
Keywords: doomer girl, pale skin, black hair, medium hair, forehead, parted bangs, black eyes,
Optional: eyeshadow, lipstick, pale skin, black choker,
Description
This is a joke trained on three images.
FAQ
Comments (7)
so what, exactly, did you do to make this work better?
I added a Wojack version
@jorkington no, how was it trained
@whothehellcares Oh sorry, misunderstood. I fucked with Pyramid's dataset quite bit, added some images, removed some, edited and retagged what was left. Then I split it into two groups; 'Good' images that were trained at 20 repeats, and 'Meh' images that were trained at 10. Off the top of my head I think it was about 1700 steps total?
Tradwife when
The model works really amazing, and accurate
By any chance do you still have the dataset?