ZIT version
Trained on 247 images.
For women in panties, use lower values (0.5 - 0.7) since most of the dataset is without panties.
For a lower viewing angle, it is better to use a landscape aspect ratio, and for a high angle, a portrait one. (tag words: low angle, mid angle, hi angle. But this doesn't work very well.)
Trained on comma separated tags.
Works on base checkpoint.
If a girl is lying upside down, try adding "in the foreground is the huge ass of a lying girl."