Coffin of Andy and Leyley characters and art style (for Pony XL)

https://civitai.com/models/281683

(Description of LoRA on civitai page):

basics it's great at doing Andrew Graves and Ashley Graves, both as child and not; you'll want to describe their outfits, and if you have them in the same scene, you'll probably want to specify eye color.

it's okay at doing mrs. graves; she's basically just an older ashley with green eyes; it's not good at doing ashley graves and mrs. graves together, because of the different eyes, it gets confused.

framed: comic style, like from the game.

bust: torso shots of the characters

by nemlei, the coffin of andy and leyley: generates in the style of the game

usage here are some examples for each of the characters

andrew graves, green eyes, short hair, pants, long sleves,

ashley graves, pink eyes, pony tail, bare shoulders, bra strap, shorts, black choker

andrew graves, green eyes, short hair, pants, long sleves, child, male child, purple shirt

ashley graves, pink eyes, short twintails, shorts, overalls, striped shirt, child, female child

mrs. graves, black hair, medium hair, green eyes, mature female, bags under eyes, office lady

lady, light hair, grey eyes, light shirt

cultist, robes, hodd up, shaded face

It can't do mr. graves at all, it can kinda do cultist leader but it's not accurate, lady is okay, and cultist is hit or miss. And gets Andy, Leyley, and their mom pretty well though.

issues The style is heavily trained into this. You can use this with other styl loras, if you want to make images of the characters, and it works well, but it takes a little bit of effort to get the white outline and limited palette out of the images.

notes I might retrain this. This one has a relatively high 0.15 noise offset; this was necessary to get the high contrast affect from the game. I trained a version without noise offset, and it was not very good.

I might do another version with more thorough tagging on the style, I intentionally left out tags for white outline, outline, and limited palette in order to enforce the style, but that might have been a mistake.

This was trained on about six hundred or so images. Likely way more than necessary, and it took a lot of manual tagging.

If there's enough demand, I might do 1.5 version.