UPDATE: A comment suggested I use Illustrious to train. After further inspection of the resources I used to train this (Nova Anime XL IL v14), I've realized I actually DID train this on an Illustrious checkpoint, NOT an SDXL 1.0 checkpoint. I've updated the title to reflect this discovery and I apologize for any confusion. I am still a lil noob :)

My very first LoRA! Trained on 35 NSFW gpose images of my face 3 female Au Ra Xaela. There is always a disgusting lack of resources for face 3 female Au Ra, which is crazy because it's the best face... So I made my own.

Trained 6 epochs, 1600 steps on NovaAnimeXL v14. No idea if it will work with other base models. Trigger words I use to get her to come out "turlina, blue skin, indigo scales, female au ra, au ra horns".

I have tested prompting different skin colors, hair color, fingernail color, lipstick color. Those all work fine. I have not been able to change scale or horn color though, so, sorry Raen enjoyers. Will also probably have to prompt in "Au Ra tail" if you want tail. I have tested adding a tail this way and it works fine, but it probably won't show up by default because the dataset images had her tail removed.

I also forgot to remove her choker, piercings and rings when I was capturing the images for the dataset. So, if you get random jewelry, just prompt them out and it should come out fine. I plan to do another LoRA soon, trained on just her fully nude form, without any accessories adding extra, unnecessary noise. But, for my very first LoRA, I'm pretty damn proud of myself.

Preview images made with random prompts from images I found here on Civit. Workflow I used was Freyja Pixel's HENTAI1911A2, which you can find here on Civit as well. I'm sure it's also embedded in the metadata of the preview images!