MadMark Style.

For some reason no matter how I train it, it seems to lose a bit of image comprehension. Has a preference toward sharp teeth due to the training data. Putting "sharp teeth" in the negative should help.

Good between values 0.7 and 1