https://www.pixiv.net/en/users/111537263 — Follow me on my Pixiv! (I may post additional images here)
In my attempt to fill in the games with Shangri-La Frontier Illustrious models on this site.. next up is Rei Saiga's older sister and leader of the Schwarz Vulf clan, Psyger-100! Obviously this is only her in game avatar model.. if they eventually show us what she looks like outside of the game, I'll do that one too! I have a few more Shangri-La Frontier models coming up after this one :)
Trigger words (appearance)
ShangriPsyger100, green eyes, red hair, long hair, hair between eyes, side braid, hair ornament, lipstick
Trigger words (outfits)
#1: white detached collar, breastplate, black undershirt, shoulder armor, gauntlets, white overskirt, black leggings, greaves, belt pouch, triangle earrings
Description
FAQ
Comments (8)
Yay another one to play with ^^
Over the next 2 days I'll be releasing 2 Rust models also (Nephilim Hollow & Shangri-La Frontier versions) :)
@Vortigraph YAAAAAAAY that makes me even more happy.
btw i have a question what kind of lora trainer do you use, i wanted to make some illustrious Lora, to be specific the Loras i already done in pony, the problem for me is that the ones i found are to confusing for my adhd brain
@wochenendenr1272 all of my LoRA are trained on the CivitAI trainer, not because it's better than local or anything, but because it's a good trainer that doesn't lock up my computer for doing anything else while training :)
@Vortigraph ah okay, are there settings you could recommend?
@wochenendenr1272 I honestly think I leave everything more or less on defaults.. but if it helps, the settings I used for this model specifically looked like this: https://i.imgur.com/JqVz8Dc.png
I train 15 epochs and usually somewhere between 7~9 repeats, aiming for around 1500~2500 steps (variance on these numbers is down to size of the dataset). I think more important than the training settings is preparing a high quality/diverse dataset. I personally try and acquire between 25 and 50 images, and I usually split the images (roughly) 15% full body, 15% back of the character, 15% profile, 15% high detail close-ups.. with the rest just general portrait shots with plenty of varied poses/expressions.
I only use anime screencaps (from blu-ray rips where possible) or official art of the character from their official website, I spend some time in photo editing software (gimp/photoshop etc) cropping all pictures to 1024x1024, removing things that will confuse the training as I go (like additional people, trying to add varying background colors/patterns to keep the model flexible etc).
Hope this helps!
@Vortigraph I helps very much thank you