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 (17)
I find myself going back to 5.0 fairly quickly with each release :( I just seem to get diminishing hits. Hits hard when it hits, but more and more seldom.
As long as you have a version you like I guess that's good! Sorry to hear you're not a fan of the progress though.
So, it's less stable for you? In terms of good outputs? What kind of outputs?
For example, maybe it's less stable with "general" outputs, but now it produces a lot better penises. So this is an important development to move into a stronger general NSFW direction.
Or is it about the style?
@6tZ Prompts that used to be stable are getting washed out, noisy, and body part melding/disappearing. Overall less realism, more distorted proportions. I'm like 1 for 5 whereas 4/5 on 5.0. What kind of prompts? all kinds, from straight realism to fantasy/sci-fi/horror all of which may or may not include a NSFW variant. I haven't sprung for 6.2 yet....and I see I've commented on 6.2's entry...oops. Sorry about that! 6.2 might be the best thing ever....
@6tZ all kinds of outputs, from nsfw to realism to genre and all combinations. But I see I commented on 6.2......to be fair I haven't sprung for 6.2...short on buzz. Didn't mean to comment on this current one, sorry about that!
@The_Meridian_435 Gotcha, yeah more testing is always appreciated, and feedback as well.
6.2 is still not stable. Again, I'm focusing on advancing the NSFW capabilities mostly, but it can definitely deteriorate anatomy. I would say wait a bit with the versions until you're sure of what you would get.
I agree. I used version 6.0 and was quite disappointed. The quality simply cannot compare to other checkpoints like red z and moody.
In future PLEASE just release FP8 or Q8 GGUF versions of all your models!!!
They are awesome but majority of us have old hardware that cant run BF16/FP16 versions :(
Thank you!!!
what do you mean "just", how about "too"? Many of us use the 16 models.
The BF16 versions usually have almost 4x more downloads then the FP8 version when both is uploaded on different checkpoints.
It's just a lot of work to support all precisions and quantizations.
The BF16 is also a lot higher in quality, so it is the priority.
When I upload them for some models, it's not often used.
But I'll do my best.
FP8 seems to be the best compromise.
@AKDesigns FP8 is uploaded.
I am trying to use controlnet (Z Image turbo controlnet union) with this checkpoint or any other, and I get really bad results. Is it not possible to use controlnet with any z image checkpoint?
"ReferenceLatent" node for "Input of Controlnet-Image". "ClipTextEncode->ReferenceLatent->Pos.Ksampler". "ClipTextEncode->ReferenceLatent->ConditioningZeroOut->Neg.Ksampler". Use Regular Controlnet like Controlnet Aux. Care Resolution. GetImageSize->PixelPerfectResolution->resolution.
Is any particular version of this compatible with character loras?
They all are, you just have to trin it on ZIB not ZIT
@mastodonlopez351 I've just started on my ZiB journey. I have something to show, but it's far from stable yet ^^
Check out the Zora model (https://civitai.red/models/2581210/zora-z-image-base-nsfw-xxx-model)