Trained on realistic images focus on POV.
best output is around 0,8-1 for me
new dataset containing 160 images
deepthroat_v2_krea2_3292840_epoch_20.safetensors
