The LoRA is a personification of Naenara Browser based on Merryweather comics.
For this one, i used 2/1 dim/alpha.
add lora in prompt, then use the trigger keywords.
Ideal weight: 1
can go from 0.6-1
Trigger: naenarachan
Base: 1girl,solo,red eyes, black hair,single hair bun,hair stick,long hair,hair ornament,fox ears, fox tail,red tail
Clothing: korean clothes, red shirt,blue skirt,pleated skirt,long sleeves,hanbok,white bow,red footwear, serafuku
The LoRA is very modular and at this point, I don't know why. All LoRA that i generate demands that whatever details you want should be put in positive prompt. I have no way to fix this man. But i'm researching to fix them....
All critics/recommendations are welcomed so that I can fix what I'm doing wrong.
Training Stuff:
Colab: HolowStrawberry's
Config:
unet: 0.0001
texenc: 0.00002
batch size: 2
repeats: 1
epoch: 10
lr scheduler: constants_with_warmup
lr_warmup_ratio: 0.1
dim/alpha: 2/1 (yes, you read it right)
611021448168921775