Trained on Hassaku, PixAI site, 12 images. First attempt, it seems to have turned out well enough.

I made some notes on some issues I ran into as well as my suggestions for solutions. It would seem to be compatible with many models on PixAI site, so far, I already used Coffee NSFW, Koji 2.0-2.1, Hassaku, Yotta Mix, GrahamAingPikas2.3, DarkSushi and etc.

LoRa Strength: So far, I personally prefer 0,6 - 0,7, you'll get a good result at higher strength too.

CFG: 6-8 (high camera angle) might come in handy, otherwise your image's point of view will likely be from knee to crotch area looking upwards.

You might want to use strong Lora for good, detailed background. Also, I suggest to put pressure on ze prompt black hair, I think most of my examples for PixAI trainer had some sort of "coloured light" illuminating her hair. I used "pure black hair inside "()", and it seems to have punched it down quite a bit, you'll still notice if you look enough, but it's not as obvious.