A lora based on a very skillful artist (dduck kong)

• Works best at 0.8-1.0 strength

• Use clip skip 2

• Easynegative embedding

• model used orange

Training settings

92images

10 repeats

4 batchsize

6 epochs

1e-4 learning rate

1e-4 text_encoder

1e-3 unet_lr

512res

128 dim

128network_alpha