Trained on 93 images dataset (x 69 repeats)

First shot without fine-tuning or reg images

with WD14 automatic captioning (un reviewed)

This model is little over trained, use better at 0.6 - 0.75