Z-image turbo has good out-the-box knowledge of breasts, but Personalization is limited to a degree.
There are many wonderful loras from creators on here so far, but all seem to manipulate the entire image, often modifying facial features and characters
Training this lora, I have found that, Myself, has created something that does something similar.

with that admission, I believe its not the fault of any creators, and their contributions are always appreciated. its something that All of us will learn how to fix in the coming weeks.
through learning and redoing loras. - done this

about version 3 (V3) (theres no V2 - the training failed)
the Third version.
trained on cropped AND RESIZED images of one of my favorite breasts shape. just an experiment, it works a little better

about version 2 (V1)
the second version.
trained on cropped images of one of my favorite breasts shape.
Doesnt have the ability to modify facial structure/features - atleast not as much as V0. I think I have reduced this by 90%
softens the images, no, it doesnt

about version 1 (V0)
the first version.
trained on cropped images of one of my favorite breasts shape.
has the ability to modify facial structure/features - so using it with characters is still problematic
softens the images, I will decide if I could fix this, or not.

I am not promising to recreate this lora. I did it 🤷‍♀️​
My workflow involves Adetailer which cleans up anatomy much better than in comfyui facedetailer, so i do not need to make this work as a ZIT lora for that