Type | |
Stats | 2,331 41,390 |
Reviews | (384) |
Published | Jan 28, 2024 |
Base Model | |
Training | Steps: 3,184 Epochs: 12 |
Usage Tips | Clip Skip: 1 Strength: 1 |
Trigger Words | score_9, by nemlei, the coffin of andy and leyley, |
Hash | AutoV2 8FCCA9C0D5 |
The Coffin of Andy and Leyley
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 charactersby 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.