--------------------- V3.0 ---------------------
A lot of effort went into this — hope you guys like it :)
This V3 release includes several major improvements. I carefully reviewed the feedback I received and addressed the issues present in the previous LoRA. Updates include:
– No more color fading
– Faces are no longer altered
– Improved genital anatomy
– Greater overall versatility
– A fully reworked dataset
– A new training method
– Support for both shaved and hairy anatomy
– More body-type diversity (not limited to skinny women)
– More consistent close-up results
Use this lora between : 0.7 - 1
recommended triggers : shaved pussy, anus, hairy pussy
Sampler : Euler / beta - 10steps
worflow embedded in each pictures
--------------------- V2.0 ---------------------
improved high res dataset
no more faces in dataset
adjusted training params
better skin texture and genitals
Recommended settings :
Lora weight 0.7 - 1
Full workflow embedded in the example images.
First try on training a z-image lora.
Keep in mind this is trained on the distilled version and not the base model (not released yet).
Once the base model is released i'll re-train and update the lora.
No specific keywords but i recommend using words like : naked,exposing her pussy.
Description
FAQ
Comments (21)
Seems to work very good. But only produces pale skin for me. Please train with more variety of skin tones :)
@ardanlilith622 Thanks! You’re right—I’m planning to rework the entire dataset to include a wider range of ethnicities, more diversity in pubic hair styles, and better representations of breasts in different sizes.
@aip0pp could you please train on a diversity type of areolas, they look all flats, ik lot of womens have it like this but its better to have pumped areolas not too much just for aesthetism.
"Crowd is very diverse (arabic and black mostly ,age, body, genre)"
wow that was interesting, i actually wonder why you typed that 😂😂
For comparison, I created a workflow where everything was the same, with only the difference of simultaneously starting with the same model and going through this Lora in the middle stage, and the result was that the overall color/body/face details of the characters were completely different. It almost reminds me of qwen.
Maybe it do somethng but i didnt noticed
Good work !! And nice workflow but i can't use the upscaler idk why
male nudity when?
This is the best ZIT LoRA I've ever used—it doesn't alter the face, and the body structure stays intact too.
Great work! The only Lora in all of Civit who doesn't affect the face.
can someone tell me how i can use this lora
hey, workflows are included in the images, just copy paste the nodes
@aip0pp how can I combine this lora with other character Z-Image lora? any workflows? should I connect sequentially? what should be the weights? also will this lora lessen the likeness of character lora?
@aip0pp i copied the images and put them in comfy ui as the nodes but nothing happened, sorry if im bothering im just new to all this
@djonjajaja201 no problem we all started at some point. you can click on an image and click on download top right.then move it to comfyui to load the workflow or click on the i at the bottom of any image and click on copy on nodes and then paste into comfyui :)
@aip0pp Thank you very much i watched a lot of tutorials the last couple of days so im way more informed now!
Nice pussy, brother!
How it's "better"? For years models generate genitals worse than hands, lol