This LoRA is designed to generate a single afterimage containing details around the female crotch area.
The training dataset consists of 30 images in total:
10 full-body images viewed from the front or back
20 cropped images taken from full-body images or manga panels
In v0.0.0, this LoRA works properly with vertical afterimages, such as those seen in missionary or cowgirl positions. However, there are no training images featuring horizontal afterimages, so this type of afterimage is unlikely to work properly.
Description
Initial.
dim16/alpha16, AdamW8bit, UNet LR 1e-04 668steps(1024² constant 540 steps + 512² cosine 128/160 steps), 30 images.