I have added three versions of reapers into this dataset. The basic reaper one, the dog strain from "Blade Trinity", and something I have decided to call a V-strain.

Tags for the basic version are as follows(if you want open mouth, if not - edit tag accordingly):

reaperponyv1, 1boy, teeth, sharp teeth, fangs, bald, split jaw, mandibles, open mouth, proboscis-tongue, tentacle tongue, dark veins, monster, colored skin, grey skin,

The dog strain works rather well, which surprised me, given the low amout of images. It also works rather well with anthro characters, even if you don't use the "dog-strain". Tags are these:

reaperponyv1, dog, animal, dog-strain, teeth, sharp teeth, fangs, split jaw, mandibles, open mouth, proboscis-tongue, tentacle tongue,

And then there's a V-strain. I have found images and photos of a different version of reaper that wasn't in the movie. Their mandibles are raised even higher and at the base of their neck there is another fang, all together they form V-shaped open mouth, hence the name. I should warn you, that using basic reaper tags - proboscis-tongue, tentacle tongue - makes it revert to the basic version, not V-strain. Tags for it:

reaperponyv1, v-strain, 1boy, teeth, sharp teeth, fangs, bald, split jaw, mandibles, open mouth, dark veins, monster, colored skin, grey skin,

Keep in mind that if you want their classic split jaw to generate correctly, you should keep the Lora at 1.0 weight, otherwise chin and lower jaw start to appear.

I was very greedy with the images for its dataset, there were many sketches, busts, and similar images, which I didn't tag accordingly - no "disembodied head" tag. I will fix it if I ever decide to make a v2.

Useful negative tags:

bust, statue, monochrome, sketch, disembodied head, disembodied,

SD is still weird when it comes to scars - even if their chin scar works most of the time, the chekpoints tend to cover a character in scars head to toe.

I also tried to do everything I can to make their back gills(or whatever the Hell those are), tried to put every image I could in the dataset... still, they refuse to work the right way.