100 images, with the training parameter:

--train_batch_size=4 \ --unet_lr=2e-4 \ --text_encoder_lr=5e-5 \