The apparently popular trick of telling Z-Image to produce translucent yogurt isn't quite to my liking and, in fact, it does look pretty much like yogurt but if that's your thing, go right ahead, just add blueberries \o/, unless you're allergic O.o. Or, hay, I have an idea, any LoRA probably does a better job.
This LoRA is intended to produce a significant amount of male ejaculate on a female subject, or on pretty much anything.
This model doesn't imbue the generated image with any particular face since there were no identifiable subjects, human or otherwise, in the training data. I've seen the phrase "non face changing" used and I could have used that here as well but I think the prior is more to the point. I have tested with various characters and the character remains undamaged.
So, effectively, if you have a character LoRA enabled then this LoRA should not alter the face characteristics in any way that obliterates the likeness.
For version 2.0 I put quite a bit of information in "About this version", including suggested weights to work with.
Description
FAQ
Comments (4)
At 1.0 it's too much plasticky and actually change the face (typical plastic ai face), but at 0.3 and with some yogurt prompting it's really good.
The image is a close-up portrait of a woman with dark hair with tiara on her head. She have streaks of sticky translucent yogurt on her face. She is wearing a pink dress with a gold necklace and earrings. The woman is looking directly at the camera with a slight smile on her face while translucent yogurt is covering her whole face. In front of her, there is a large pink letter "N" that is lit up with white lights. The background is a soft pink color with a subtle texture. The overall mood of the image is romantic and dreamy.
Subtitled: "For those seeking the Crème de la Crème!"
Hey I don't know much about loras so this might all be idiocy but it looks like version 2 might not work with the new nunchaku'd z-image turbo model for some reason - version 1 works fine, applies lora to all modules, but when I run version 2 it says
"[Z-Image LoRA] Applied 0 modules; check module-path mapping and LoRA key formats"
Of course the z-image lora loader hasn't even been added to the custom nodes yet, so it might just be a bug, but it has run every other ZIT lora I've tried so far so I thought I'd give you a heads up, because I'm a huge fan of your work.
Hay \o
Thanks. Now that you mention this I think it's because v1 was made with ai-toolkit, and a different adapter, and v2 is definitely OneTrainer. So I don't know what that's going to mean for my other LoRA but I've moved over to OneTrainer and most of my ZIT are made with that.
So I'm not sure what to do with that information but thanks, and I, myself, tried Nunchaku for Z-Image and it wasn't ready for prime time and broke my Qwen setup and I didn't really need it so I never figured out how to get it working, and yea I think it wasn't completely ready yet. Though I wonder if the rank could have something to do with it, and if that's the case then it's a goofy bug with Nunchaku and we'll just have to wait.
So thanks for the heads up, cheers! \o