A Lora based on a character from Counter:Side named Seo Yoon. This is her "awakened" form. Ver 1A and Ver 1B. Lora is made by cutting up images of the said character from to 1024x1024 image dataset training. Trained from 60 images with 6000 steps. Feel free to leave a comment, It will be appreciated! V2 is completely revamped to increase versatility while the quality did get reduced, versatility is definitely more important in most cases. And I actually forgot some of the tags, I made this like a month ago and I forgot to post it here LOL. V1 IllustriousXL is adjusted to suit IllustriousXL base.
Feel me to support me at Madafada1991 is creating SDXL lora for AI generative image. (buymeacoffee.com) Any of your cent will motivate me to make more LORA!
Standard outfit prompt: black suit outfit, boots,
Dress prompt: black dress outfit, cleavage, underboob, bare legs,
Bloomer prompt: bloomer outfit, sporty boots, loose socks, bare legs, t shirt, ponytail,
Ver 2 is tuned to be extremely versatile and good at NSFW
Ver 1A is a fully tuned to maximize similarity to the character look in game, with very high accuracy. (Recommended)
Ver 1B is tuned to be less accuracy and have more freedom to design her using prompt. (I mean why not? almost similar capability with more freedom?, recommended to get this too!)
Hassaku ver is cut down version of the other two, with some dataset removed.
Credits to
CagliostroLab Creator Profile | Civitai
For making such an amazing model!
Description
As stated, cut down further version, best used on 0.8 strength.
FAQ
Comments (11)
Could you make this in sd.1.5? Any anime model xD
Theoretically i could but no matter what I could do, SD 1.5's limitation is 512x512 pixel, so there is bound to be huge data losses, so I might even need to change the dataset as some data is bound to be disturbing the color and consistency of the said Lora. I tried a slightly different dataset a few months ago on a lot of famous SD 1.5 models(like anything v4.5, hassaku and meinamix, where most of the problem will begin in body lower part), and the best I could do is a subpar lora(in 2024). You could say that they are obsolete sadly.
In case you want a speed and fast model for SDXL why not try out VXP Turbo 1.5? You could in theory ran this Lora with that checkpoint model.
@Madafada1991 I understand. I can’t run SDXL sadly but would if I could.
Your Loras looks so good though, too bad I can’t try it out xD
You could try train it in 1.5 using SDXL settings :p
@Shunin i made it for hassaku. you can check it out.
@Madafada1991 Awesome! Thank you so much!
Thanks for making counterside character!! <3 <3
no problem!! <3
it looks really amazing though.
thanks! appreciated!
Could you please make Pony version?
I lose her training data unfortunately after reinstalled my OS. so it will take some time to make one.