Meet Minnie.

A face/identity driven LoRA built off very tight crops on a very defined data set.

Kitten's partner in raves, fun, crime, and long time friend.

Works hard by day, works harder when playing at night. A friend you love on your side but would hate to be on the wrong side of fury. Don't let her name fool you, she's the more ferocious of the pair.

What does it reinforce?

A more defined and prominent nose, even when pushing to anime models, a wider defined bridge, prominent lips, long eyelashes, defined eye shape and brow, well manicured and thicker eyebrows, high defined cheeks, and a delightful pull to olive and darker skin tones when prompted (with a darker base than Kitten). It should blend well in to most Illustrious-XL models and merges, most SDXL models, and plays nicely with many Pony-XL models. It is also heavier baked than Kitten's LoRA as the face itself has more flexibility and less requirements to pull out specific sub-features, so it plays better out of the box with Illustrious models that focus on realism.

The hair color streaks are baked into the LoRA, you can absolutely output without them but higher LoRA weight will more strongly define them to the original base.

A better focus on realism than my last launch and many anime models will play nicely, but the less nose definition in the general model the more you may have to push the LoRA.

Usage

• Trigger Word: m1nn13f4c3

• Prompt Helper Words (for original character design)

  • dark hair with bright blue highlights

  • dark hair with bright blue streaks

  • olive skin
    (NOTE: olive skin works fantastic on most models to give her the nice base color, but if using the on site generator without additional details in the prompt it is a coin flip on normal skin or looking like She Hulk, when using this as a trigger/helper, add additional details!)

• Illustrious Line (General)

  • LoRA Strength: 0.6 to 1.2

• Illustrious Line (Realism Focused)

  • LoRA Strength: 0.4 to 1.0

• Illustrious Line (Anime Focused)

  • LoRA Strength: 0.4 to 1.3

• Z-Image-Turbo Line (Sketch/Artistic)

  • LoRA Strength: 0.7-1.2

• Z-Image-Turbo Line (Realism)

  • LoRA Strength: 0.4-0.6

As much as I hate the full l33tsp34k version of the trigger word, trigger styling follows Kitten's face/character LoRA, which was intentionally set to alphanumeric to avoid overlap with unintentional furry output on multiple anime models.

Why does Minnie Exist?

Since starting AI generation coming from a background of photography and design work, the randomness has bugged me, especially when I find a specific character out of a large batch I like, then spend hours (or days) trying to chase that specific set of features again on other seeds, with other models.

Throwing mountains of prompt tuning at the problem, breaking model token limits constantly, only getting truly "close" a few times out of hundreds of images, but rarely ever finding "it" again.

So, I went all in on learning how to train my own LoRA, painstakingly worked through hundreds of Qwen Image Edit workflow runs that exceeded my VRAM, prompt tweaks, outputs, on and on. But in the end, I found her again, and again, and again.

Success.

Training Information

The dataset that drives this is very narrowly focused on the face, so it should play nicely across almost any body type, hair styles, hair colors, and most clothing/outfit LoRAs that don't overly pull on or push face characteristics. Trained on 65 tightly cropped face photos from the same source image, it should hold the character's primary identity while allowing for near complete customization everywhere else.

Baked heavier than Kitten's LoRA to pull harder to skin tone, hair styling, hair color streaks, and moderate shoulder tattoo deployment. Should still allow for flexibility in all of the above, but a reduction in weight may help hard shifts from the default character.

Acknowledgement

This LoRA could not, and would not, exist without @purplelady's Elegence Niji LoRA (oddly enough, the Z-Image-Turbo version) creating the original style I found and loved, and the original version of the character that came out of it. You'll find as a result that due to that core structure it play's wonderfully together with the Illustrious version of Elegence Niji, and I highly recommend it if you're looking to amplify and reinforce this specific set of facial traits in your output.

The the source and training data is many steps removed from those original Z-Image outputs, the concepts that lead to this chase never left.

Next?

The ZIT version of Minnie's LoRA, enhancements to the original Kitten LoRAs, and more friends to come.