Trained with different parameters, should be more consistent with cutout combined with non-standard prompt (like skirt or pelvic curtain) and poses (like kneeling or from below). The cutout shape might be less flexible though.
Dataset (including captions) didn't change, see previous version.
Trained with different parameters, should be more consistent with cutout combined with non-standard prompt (like
skirtorpelvic curtain) and poses (likekneelingorfrom below). The cutout shape might be less flexible though.Dataset (including captions) didn't change, see previous version.