This model can produce realistic selfie and smartphone quality images.

Optimizer: ADAMW
Learning Rate Scheduler: CONSTANT
LR: 0.0001
batch size: 3 (vram limit)
Learning Rate Cycles: 3
res: 512
LORA rank: 128
LORA alpha: 128

dataset: 113 images
GPU: rx 7900xt
tool: Onetrainer