My first attempt at creating a lora containing both a style and characters feedback is appreciated, I used 200 total images from the show, 75 of them being random characters, scenery, etc and 25 for each character, overall I think it turned out alright but I will train it again because some of the characters aren't consistent and sometimes the style isn't as visible. trained here on civitai using as a training model at 1024
Style info:
trigger word: kan_style
weight: 1 is perfect
Characters info:
Trigger word - looks tag
Kim: kimaps - long hair, orange hair, green eyes
Monique: moniiqkue, black eyes, black hair, dark-skinned female, dark skin
Bonnie: bomnkinie, short hair, brown hair, green eyes
Ann: ankkiin, brown hair, short hair, blue eyes, doctor, lab coat
Shego: shmeikgo, black hair, long hair, green eyes, gloves, bodysuit, lipstick
Additional information:
I use
I use
as an Upscaler
Imagine dimensions vary depending on what I'm trying to do
Sampling steps I go 20-30
hires steps I go 8-15 (I don't always use hires fix)
Denoising strength I go 0.4
I use 7 CFG Scale and set at 9
for faces, I use mediapipe_face_mesh set Dilation to 50 and Y offset to 30 to get the full face
I do all of my lora training here, if you can spare any buzz I would appreciate it