The best or only "evil" secretary in Nadesico. My second favorite character of the series, the first being Inez Fressange and probably the only other good one is Yurika Misumaru. Getting images was a pain and ended up using mostly screencaps which i had to scrub and filter due to graininess (stupid 90s low res crap). Anyway it ended up ok-ish. SD seems to really hate lipstick for some reason.
V2:
I was a bit displeased with the V1 LORA and as I am doing some test with the prodigy optimizer I decided to do a retrain. It looks a bit better and is more stable.
Got ok results with: <lora:ErinaKinjoWonV2:.7> ErinaKinjoWon
I tweaked her outfits tagging to make them a bit more reliable the new tags are:
nergal_military_uniform_black_pencil_skirt_grey_thighhighs
erina_red_skirt_suit
V1:
Got ok results with: <lora:ErinaKinjoWon:.75> ErinaKinjoWon, large_breasts,
Available outfits are:
NergalUniform
ErinaSkirtSuit
Warning: SD seems to hate lipstick, when i find a fix i might do a V2.0. It seems unwise to use lipstick, red_lips or makeup in the NAI family models. Anyway that issue lowers the success rate of this lora. You can just say screw it and add lipstick and red_lips in the negatives to nip the issue at the bud. After doing a stupidly huge xyz(I'll add it as an extra gallery) plot I determined you can use red_lips and lipstick at no more than .55 weight combined so (red_lips:.4), (lipstick:.4) are good values to get decently looking lipstick.
Might be a good idea to add skirt, miniskirt or pencil_skirt to the outfits as sometimes it likes to give her pants(Which is not her cannon outfit.). Canonically she uses either stockings or thighhighs.
You may also want to add short_hair as in some front shoots the lora seems to like adding her long hair or a ponytail.
Her mole seems to come and go with the LORA strength. Use mole or mole_under_mouth at your own peril as the model(anythingV4.5) may go way overboard with the amount of moles.
Since this was mostly trained with screencaps(Don't want to be sued :P) i will refrain from posting the dataset, in it's place I'll add the tag summary.