*** Any feedback on how I can improve the LORA would be greatly appreciated***
Skullgirls is one of my all time favorite games. I've put a stupid amount of hours into it and I love the game's aesthetic. Been wanting to make as close to a 1:1 style Lora for the game for awhile now and while I wouldn't say it's exactly perfect (colors are bit of a challenge to match exactly) I think I've gotten pretty close.
v1: Smaller file size, more flexible, but not 100% as accurate as v2 (still overall pretty accurate to the game's style).
v2: Larger file size, way less flexible than v1, but more accurate to the game's style.
I suggest using a .70 weight.
If you're struggling to get a specific hair color to gen or clothes use emphasized weights as needed. See image prompts for examples.
activation prompt: sgtyle
suggested negative prompts if you don't want everyone having FIlia's hair or Parasol's dress: extra mouth, prehensile hair, detached hair, short dress.
If random buildings are being generated in your background neg prompt: city, buildings
Big thanks to Luna-Chan for all their help figuring out training parameters.
Description
version 1
FAQ
Comments (6)
I don't mean for this to sound like "It has no diversity" but: It's very good but it struggles to generate PoC. I use the same prompts with and without this LoRA and while without works, with it kinda struggles and makes it a non-human skin tone. Overall very good but I hope to see it have more variety, kinda like Fukua.
PoC?
@tokoat People of Color
I think V1 is better at making darker skin tones.
From my experience most "artsy" models will struggle with brown skin, since they probably weren't trained on art featuring black people. Most of these styles will be based on datasets from Asian artists who draw mostly pale-skinned characters.
HI, WHAT IS THE CHECKPOINT YOU USE?