This LoRA was born mainly from my curiosity to create one, to understand the training steps, and to dive even deeper down the "rabbit hole" of this fantastic world!
The primary goal was to "fix" what I consider a "flaw" in many Z-Image Turbo models: breast sizes are often standard (or rather, large). Very often, the prompt alone isn't enough to get the desired size, and it’s difficult to achieve a small or very small bust (at least, that’s been my experience).
As a long-time fan of small breasts and a recent fan of Z-Image Turbo, I thought this was the perfect opportunity to test my first LoRA! And while I was at it, why not try to fix the "nipple detail" issue too? After all, it’s just my first LoRA: how hard could it be to do something simple? What could possibly go wrong?
Well, no need to beat around the bush: something went decidedly wrong 😂😂
The dataset was basically entirely based on small breasts. You can imagine my surprise when I realized it did everything except change breast sizes! But not everything went wrong. I noticed that part of the initial project actually turned out quite well. Right from the first generations, I could see the texture, skin details, and lighting improve compared to generations without the LoRA (obviously, this is just my personal taste).
So, even though I was disappointed that my primary goal wasn’t achieved, I’m pretty happy with the result: half the idea worked! And anyway, it’s just the first experiment; maybe the next one will go better.
Anyway, you probably don’t care much about my reasons or my fantasies about small breasts... 😉
I’m publishing this LoRA because I think it could be useful for users looking for a slight improvement in textures and lighting, or for images that look more "professional." If you are looking for a raw "amateur photo" look, this isn’t for you.
It doesn’t pretend to be a magic wand, nor to replace existing LoRAs that do this job better. However, it can be that "utility" LoRA that fixes sma