XL version of my simple FaeryQueen Lora.

Trained with anime style images, but apparently can work okay with realistic models too.
(Training images at https://huggingface.co/datasets/ppbrown/faeryqueen )

Even though anime style art was used, it was trained on sdxl base.

Note that for some odd training reason, putting "red eyes" in the negative, often normalizes things.

Funny thing: I noticed that this dataset in my SD1.5 Lora sometimes gives red eyes, so using it there was somewhat obvious.
For this lora.... I dont see red eyes, but putting it in the negative still seems useful.