Miyuki Ayukawa from the anime series Basquash. Another LoRA training experiment this time using blurry low res screencaps taken from an ancient streamed release.

How to use
Depending on model and what else you're got running anything from "bsqam,1girl" to "bsqam, 1girl, dreadlocks,twintails,breasts,purple eyes" might be needed.

Breast size wasn't pruned from the training data so whatever default size you get is up to your setup.
"Bandaid" and "bandaid on face" wasn't pruned either meaning they generally won't show up unless specifically prompet.
Keep in mind the animators mixed up which side the bandaid should be on on several occasions so if you do you might get it on the wrong cheek, both or somewhere else on the face.

Seems to retain character appearance fairly well while at the same time adhering to model and prompt style influences.
See included sample images for recommended settings.

Misc
768 resolution trained on NAI without noise offset option for best LoRA compatibility.
Tested on NAI + AOM2. All sample images are raw untouched txt2img except main logo.

Dataset tag occurrence
bsqam 100%
1girl 93%
bandaid on face 80%
solo 80%
bandaid 80%
twintails 67%
jacket 60%
breasts60%
cleavage 60%
crop top 53%
midriff 40%
sky 33%
open mouth 27%
medium breasts 27%
smile 27%
navel 27%
belt 20%
large breasts 20%
from behind 13%
facing away 13%

Model Information

Miyuki Ayukawa [Basquash!]

Model Details

Available Files

Tags

Versions

Related Models