Z-Image is here, and it's pretty damn great. Let's make it more NSFW!
Still a work in progress. It can output good genitals, but not always. About 3/5 images are good, working on penis stability specifically for future versions.
ComfyUI: Models should be placed in diffusion_models, and uses the official workflow with the Load Diffusion Model loader, and loading the VAE/CLIP models separately. Models named "_AIO" should go in the checkpoints folder, and uses the normal checkpoint loaders.
Recommended Settings
Steps: 12
CFG: 1.0
Sampler: dpmpp_sde
Scheduler: simple
Description
FAQ
Comments (39)
What's new in 6.0? Also, BF/FP....not sure what's the difference there? Thanks.
Different optimizations for low VRAM or quality.
FP16 is worse quality but may run on lower hardware.
6.0 improves NSFW in general. A new muted merge of some NSFW models are included.
Next version has a penis improvement pass.
it's quite dangerous...
What about penis in 6.0 model?
Always aiming at improving the penis. It's just super hard to get it to do well on both male and female genital aspects :(
6.1 will focus more on penis.
why most of these checkpoints are overwhelmed with female photos? is it not capable of creating male figures?
Why spam this under every checkpoint?? Ask once and otherwise check the galleries. Most checkpoints can create both men and women.
"y r u gay?"
@SaburoDio No, we just want a complete body, a complete penis.
If you had asked, "Hey I am trying to do normal hetero couple fun stuff, will this model work for that anatomy-wise?" you would have seemed more coherent and less terrified of attractive female nudes.
Also, "we" are you royalty?
with all due respect...
@Walnutphysics I don't care if they prefer the buns or the weiners. But this seems like a bot (recently created and spams the same comment and never replies when he gets an answer).
@cai567890 ok, then you can be the tester for all these checkpoints and let everyone know if they are capable. That would be more productive than spamming complaints
what if different between bf and fp??
FP is a quantized version for those with low VRAM storage. People say there’s no difference usually but that is not true. BF = better prompt adherence and quality since it has more knowledge
@freestuffpl0x42069 Thank you
@zelot111210 No problem. If you want to know the technicals NVIDIA made a pretty good read on it https://docs.nvidia.com/deeplearning/tensorrt/latest/inference-library/accuracy-considerations.html
Atm my favorite quality women are created by dafZIB 10-15steeps and Fusion4.5 7-8 steps. really good quality and composition. Waiting for 6.0 to give it a try.
This is an excellent work, except for the slightly warm colors. It is the best model I have used in terms of overall performance.
Nooo say it ain’t so. I’m so tired of the saturation = cinema. I’m glad the prompt adherence is there at least, it’s probably because of the bf16 if you’re using it. Two others you’d also like would be Stable Yogi’s ZIT, and Unstable Revolution. Those also use bf16. I think that Real Dream ZIT might also
The saturation is a simple fix, use a LUT to modify the final color profile. All my examples so far use the Kodak Portra 400 LUT, with added noise grain.
@clueless_engineer Good looking out! I appreciate that tip, I just got down rocking the amateur style so I’ll have to use that as the counter. It’s interesting to see how different people define realism. Some prefer cinematic with the bokeh and shadows while others go with anomalies, blue tone for fade, and sharpness in texture. I personally prefer the latter due to liking full body shots, but the former is good for closeups.
can anyone help me out with a workflow for this? 16gbvram
Image to image or text to image ?
img2img@sehdiishmohamedov727
@outthetrash818 @sehdiishmohamedov727 +1🙏
My workflow is identical to any other z-image (or, really, any txt2img + upscale + refine) workflow. ComfyUI has native workflows that will work fine.
Moody just released an I2I workflow (https://civitai.com/models/2444491/moody-i2i-image2image-zit-zimage-turbo-simple-workflow). i have never used it because i don't do I2I, but I have used Moodys regular T2I and they are awesome so i wouldn't doubt that this I2I is just as good.
The hair quality is horrible in 6.0. @6tZ double check with 4.5, it has the best quality. the longer the hair, the worse it gets in 6.0 bf16_diffusion.
Show me what you mean. My images appears fine with 6.0, and 4.5 is such a huge step back for NSFW.
Funny you should say that because I already skipped 5.0, after comparing it to 4.5 (I´m talking amateur candid style quality only). Still going to try it out of course.
true, it feels like each version gets a bit worse 😅 although i think i've mastered v5.0 with either euler beta / 14 steps / shift 6 or 2s a RF beta / 8 steps / shift 2
Alright, I hear the feedback. I'll look into this more. But could you please show me examples of the longer hair, the worse it gets?
@6tZ Hey man. I would love to help you out, but I for one didn´t notice that. For me it was just a general small, but noticable decline is realism/candid look between 4.5 and 6.0
Edit: I´m not the one criticizing the hair in 6.0, but I just ran a test to compare the hair of 4.5 to 6.0 and actually it is a huge improvement to my novice eyes. I also compared the skind details....another huge step up.
I don´t have 5.0 anymore, so I can´t show you what I meant by less cinvincing realism, but no need to check, because overall I think 6.0 looks fantastic and I will stick with it.
@fanboytunes0789 What are you talking about? 6.0 has a strong contrast, as in version 5.0, the best model for realism is 4.5, anything higher than that doesn't appeal to me, and 6.1 also has issues with anatomy, so anything higher than 4.5 is disappointing.
i also don't have the hair issue, it's just that the overall look becomes more and more artificial. but i guess it's personal taste at the end. i'm switching back and forth between 4.5 and 5.
@chgrau89578 5 version is already highly contrasted, version 4.5 is the best in terms of realism and color, and there is no oversaturation
@soyv4 I didnt say anything about overall realism. All I did was directly compare close shots of the hair and the skin details and it is very obvious that 6.0 is stronger than 4.5 out of the box.
the trianning tool on civitai gose wrong to train lora with zimage turbo model. can i ues zimage base trainnig?plz give me some advice
Yeah it doesn't work too well, the civit tool. I train with ai-toolkit.

