This is the SDXL version of my previous LoRA of Eureka. It should be more precise and flexible. Compared to the 1.5 version, I added more images to the training dataset, including hentai and generally low quality art which helped to generalize the character better. This LoRA was trained on the base SDXL 1.0 model so it should work on virtually any fine tune except those that diverged a lot from the base.

The main tags used are eureka, short hair, hair ornament, long sleeves, dress, purple eyes, boots, hairclip, aqua hair, thigh strap, white footwear. Not all tags might be required but they help to reinforce some key character elements. I deliberately did not train the 1girl and solo tags because they usually learn the "native" model style, also many other loras train them so it'd be hard to use them together. It's already hard with Regional Prompter.

There's a known issue with a sudden hair between eyes that happens on some models (probably a prior knowledge). Since SDXL generally doesn't play well with negatives (unlike 1.5) this issue should rather be fixed by putting forehead in the positive prompt.

I suggest using ADetailer to improve face details and regular inpainting if needed.