https://mega.nz/file/nUlGDaTb#hBSXVledAdwHIafJYfS_aceyvSRI5g4LSTjsfZ7vQ58
Himeno LORA
Let me know how well it works.
Images generated with 0.6 - 1.0 strength
#Chainsaw man Himeno L.O.R.A.
1 messages · Page 1 of 1 (latest)
whats ur settings
for what?
20 images repeated 5 times. i have a strong assumption that the file that is create is over written if you run the training repeatedly so it can mess up the embedding if you train off a different set of images
@serene fractal have you tried it btw is it any good?
batch size 4 otherwise the rest of the settings are default
rtx 3080 for gpu let me know if you need any other info
how about the captions
how'd you caption them
wd tagger
wd14-vit
0.15 strength
eyepatch, black hair, necktie, black necktie, formal, 1girl, suit, shirt, short hair, black eyes, looking to the side, black jacket, collared shirt, jacket, closed mouth, indoors, looking at another, upper body, white shirt, 1boy, black suit, artist name, bangs, twitter username, looking away, expressionless, blurry, portrait, parody, solo focus, watermark, dress shirt
here one of them
1girl is enough some time need eye patch
pt?
what is your it/s during the script?
curious since, run a 2080 ti, just curious for the diff
since 3080 similar fp16 to my 2080 ti, mem wise as well
says 1.83 in the screenshot
you can also check my red black art style lora for another screenshot over a longer period of time @crude flume
ah
yeah my bad, im dumb asf for that one LOL
