Here's Mai, it was a pain, a huge amount of images and a good 90% don't really look like Mai. Spent a a lot of time filtering them down a manageable level and then her outfits can be rather finicky. (If you use dress on her red outfit it will try to add a back counterpart to her pelvic curtain/loincloth) It just needed an insane amount of tweaking and several retrains.

Works fine with: <lora:ShiranuiMaiV9:0.7> shiranui_mai_high_ponytail,

Had to do some shenanigans adding her hairstyle to the trigger for stability.

Only trained 3 outfits:

• red_loincloth_halterneck_plunging_neckline_obi_rope_sleeveless_back_bow: He og outfits it is likely it will ignore the weird balls connected to her backbow

• black_jacket_crop_top_black_pants_belt_plunging_neckline: another more modern OG outfit

• white_camisole_pink_dolphin_shorts_short_shorts: I don't know where this comes from.

• folding_fan: her fan (not sure if these work fine)

• folded_fan: the closed version of her fan (not sure if these work fine)

• holding_fan: actually holding the fan (not sure if these work fine)

The fan tags i am not sure how well they work as i didn't care too much.

As this is mostly fanart I will add the dataset with the tag summary so people can experiment, inside there are couple images of her cow_print bikini so it be possible to invoke them but they are not explicitly trained.