Elizaveta from the popular "yes" gif and associated youtube video

has a bias towards putting her in front of a floral wallpaper background because of a lack of availability of training data. maybe i could try manually editing it out in a bunch of them, but it is what it is

if outputs have a poor resemblance try:

1: "(low quality, worst quality:2)" in the negative prompt along with a typical photographic negative prompt

2: dynamic cfg extension and set cfg higher, Mimic scale: 8, Threshold percentile: 90, Mimic mode: Half Cosine Up, Mimic scale minimum: 8, CFG mode: Half Cosine Up, CFG scale minimum: 4

3: DPM++ SDE Karras sampler

trained against lofi 2

SHADOW WIZARD MONEY GANG🗣️🗣️🧙‍♂️🧙‍♂️🧙‍♂️💸🤑🤑🤑

WE LOVE CASTING SPELLS 🗣️🗣️🗣️💯💯💯