Z-Image-Turbo - Oppai & Nipple Remake
A very simple breast & nipple improvement LoRA for Z-Image-Turbo. The base model often gives unsatisfactory breasts and nipples that aren't quite right, so I trained this one with the natural, slightly perky shape I personally prefer.
LoRA trained based on the Turbo model can affect the default face shape of the Z model. However, by using precise prompts, it is possible to get closer to the original internet celebrity face.
Recommended weight: 0.8
It was originally just for personal use, but the result turned out decent, so I'm sharing it.
I usually generate with: topless woman with medium-sized, round breasts with small, pink nipples.
Filename: Z-Turbo_Oppai_Nipple_Remake.safetensors
Z-Image-Turbo 胸部优化 LoRA
一个很简单的 Z-Image-Turbo 胸部&乳头优化 LoRA。 原模型生成乳房和乳头经常不太理想,所以我按照自己xp,训练了这个 LoRA。
基于turbo模型训练的lora,会影响z模型默认的脸型,可以通过精准的提示词来靠近原本网红脸。
推荐权重: 0.8
本来只是自己用的,后来发现效果还不错,就整理出来分享。
我通常会使用:赤裸上身,乳房圆润,粉色乳头来呈现lora。
Description
FAQ
Comments (23)
老哥,我训练的lora跟这个不兼容,加了角色lora,角色都不太像了,而且有时候会出畸形图。你遇到这个问题吗,角色lora我是用ai-toolkit训练的。
评论区里有lora角色,你看凡人的女角色正常,我自己训练的lora人物也正常,你试试oppai权重低一点0.6,然后你的人物lora权重0.8或者0.9再试试呢
经过多次测试,发现还是会影响到脸部、皮肤、氛围、光线等元素的生成,尤其是脸,如果不特殊强调脸型的话,会比原版生成的脸丑很多
是的,turbo模型会有这个问题
期待添加更多素材的2.0
can you tell about dataset how many images and steps been done here?
I only used 20 images for the dataset, and trained for 3000 steps. Z-Image-Turbo fits super quickly with such a small set.
轮廓是对的,乳头是对的,颜色是对的
但是权重高了会影响到脸型、身材甚至背景,建议权重0.7
总体评价为夯









