Trained on frames ripped from a short animated video of a close up face with sparkling opal skin.

It may not be the most flexible - was more of an experiment to see if it would take. The concept does come through a lot of the time. Perhaps this can be used to generate a dataset that would make a more flexible model.

The same tags were used for each frame: realistic,black lipstick, iridescent opal skin, close up, 1girl, makeup