Not my first LoRa for SDXL, but the first one I think I'm happy to share.

Trained on 45 images and a lot of tags, it works well with heavy tagging or simple prompts (though better with simple prompts)

use "????, nsfw" in negative or you will get boobs and genitals

Use without refiner cfg 5, weight 0.8 to 1, but you get more ethereal if you lower the cfg and more hard lines if you raise it, minimum around 3 max around 8, but you do you.

I get the best results from a merge im using, but it was trained on base 1.0 with 0.9 vae, so should be pretty flexible in terms of model usage.