Toudou Yurika (Aikatsu) LoRA

02-05-2023/V5 Update:

• Reduce the model file size to 9MB.

• Use a way lower network dim (8 instead of 160) and Alpha (1 instead of 160) to prevent overfitting

• Add a new token for each canon outfit (using booru token was a bad idea):
  GothMagicCoord

  StarlightAcademyWinter

  CasualDress

This is a LoRA, this mean that you need to update Voldy/AUTOMATIC1111's webui to use it (use git pull) .

That LoRA does not have a particular trigger (except the default <lora:LoRaName:1>).

Weight:

V3 :

I mostly tested it at weight 1.0 and it work well between weight 0.7 and 1.0. But the more you reduce the weight the less her canon outfit are accurate.

Recommended prompt:

That LoRA can reproduice some of her canon outfit.

Goth Magic Coord (purple/blue gothic dress dress):

purple_dress,drill hair,mini top hat,gothic lolita,  idol,bat wings, white ascot,frilled dress,purple headwear, striped pantyhose,  vertical-striped legwear, uneven legwear,   black wings, wings

V5:

Add : GothMagicCoord

Starlight Academy uniform:

starlight academy uniform,white skirt,miniskirt

V5: Add : StarlightAcademyWinter

Casual dress:

pinafore dress, green dress, white shirt,frilled sleeves, red ribbon, neck ribbon, glasses, black-framed eyewear

V5: Add : CasualDress

Training settings:

101 images, 10 repeats,7 epoch,7070 steps
Learning rate: 1e-4
Text encoder learning rate: 5e-5 
Unet learning rate: 1e-4 
max bucket resolution: 512 
clip skip: 2 
dim/alpha: 160 
batch size=1 
Base model: AnythingV3, this mean that this LoRA work on any model derived from that particular model like AnythingV4/4.5, Pastel Mix, AOM 1/2,etc...

Model used for the example:

V3:Pastel Mix, AnythingV4.5 and AbyssOrangeMix2 nsfw .
V5:AnythingV4.5

That character owned by Bandai Namco and Sunrise.