This LoRA is intended to recreate her looks. I haven't tried using prompts with different hairstyles since my main goal was to capture her likeness above all.

I saw another model of the same character, but the style was heavily burned into it. It was also super chunky in size—100MB LoRAs are already pushing it, 214MB is way too much, and anything above 300MB is a definite no-go. So, I decided to make one myself. 54MB should be the standard. Also, no, more data doesn’t mean better quality. This LoRA was trained on just 18 images. Curating data is more important—quality over quantity.

Surprisingly, this LoRA is quite flexible, even at strength 1, so try experimenting with everything. You can test different outfits too. I've noticed that well-made character LoRAs merge well with different styles, while poorly made ones tend to break when styles are added. That’s going to be my benchmark moving forward. I suggest others test this too. Overtrained LoRAs might be the issue—too many weights, and they start learning styles as well. While some might see this as a good thing, I personally don’t. LoRAs should be flexible from the start, so that’s my goal for all my future projects.

Tags are in the description. I highly recommend using Civitai Helper—it makes activating LoRAs way easier. One click, and you get all the activation tags, no need to type or copy/paste. I also encourage others to upload their tags in the LoRA information section instead of just leaving them in the image description—it’s not productive at all.

Illustrious might just be the best model on the market right now. Pony was a nice upgrade, but Illustrious does everything better. Very happy with how LoRAs are trained using this model.