A lot easier to use than the last version. Trained on about about a hundred images for each concept, for 20 epochs.
" on face" is the main tag with the following concepts trained on it:
These are roughly in the order of largest data set to smallest. And you should, for the most part be able to combine tags.
on face: might do any of the below
pov: can do non pov, but it's better at pov
cheek poking: poking into the cheek; should dimple the cheek
cheek press: pressing against the cheek, should dimple the cheek
over one eye: covering only one eye
between eyes: on the center of the face
on mouth: tip on the mouth, or chin, if the mouth is open
facejob: helps make the other scenes more active
pressing and poking: it's trained on sfw version of cheek press and cheek poking to help it understand dimpling the cheek
2girls: 2girls, works well with cheek press and facejob, and oral,
3girls: a bunch of different ways, combines well with oral,
4girls: can also work with 4girls or 5girls, or more, but the data sets for those were tiny, so it's hit or miss, but looks pretty good when it hits,
over eyes: not pov, over eyes, works-ish, but mainly included to make the other tags work better
reinforcement: a random selection of images were copied to a different concept as a test with the on face tag repeated, if you're having difficulty with the lora, trying increasing attention on ( on face)
on tongue, on tongue, open mouth, uvula: I didn't train it specially on these, but because of the data set, it's pretty good at it.
Trained on NAI; all the example images are on AOM>
