2000 steps, 30 training images, 20 repeats. The model could be better, and it could use some better training data.

bf16

Base model: runwayml/stable-diffusion-v1-5

Optimizer: Adafactor

Learning Rate: .0001

TE LR: .00005

Unet LR: .0001

Network dim & alpha: 91

I recommend using both a face and eyes ADetailer as the eyes tend to get 'squinted' just running it through a face detailer. Add 'eyelashes, big eyes, brown eyes, eye shadow' to the eyes ADetailer prompt for best results.