Jib Mix Zit is an upgrade for realistic images, sharper and prettier faces.
Jib Miz Zit V2:
Better fine small details (especially in backgrounds)
More variation in image and Style (Doesn't always produce photo realism but can be a nice surprise)
Less Zit mottled Skin.
More Varied faces and poses.
- A bit less photo-realistic.
I recommend the FP16 model for 12GB of Vram and up
But a smaller FP8 model is available.
For Best Image Quality try:
I recommend using 80% of the UltraFlux VAE merged on the fly with 20% of the Normal Zit/Flux VAE for sharper images
UltraFlux VAE: https://huggingface.co/Owen777/UltraFlux-v1/tree/main/vae
This custom nod: https://civarchive.com/models/2231351?modelVersionId=2638152 does a merge of the 2 VAE on the fly as the UltraFlux- VAE can be a little too sharp-looking on its own and definitely is if you use it for upscales.
V1 model can benefit from a small amount more of my Zit lora (around 0.15): https://civarchive.com/models/2194714/jib-mix-realistic-z-image-lora
I recommend ClownSharkSampler and Linear/ralston_2s for detail.
with some added noise option nodes.
It has been tuned against my workflow (I have a version posted here: https://civarchive.com/models/2194714?modelVersionId=2481800)
Description
Better fine small details
More variation in image and Style (Doens't always produce photo realism but can be a nice surprise)
Less Zit mottled Skin.
More Varied faces.
FAQ
Comments (16)
does it works with loras we trained
Yes it works fine with lora trained on original ZIT (I just train all my loras on ZIT), well as fine as ZIT works with loras, which means you usually have to lower there strengh to between 0.20-0.50, especially when mixing loras.
Do not use ultraflux vae creating artifacts and noise https://imgsli.com/NDM1MTI2 zoom in to see the artifacts, noise, an image shift
Or use it at 40%-60% strength with this node: https://civitai.com/models/2231351?modelVersionId=2638152
and it looks perfect! :)
I use just the base ZIT/Flux VAE for Ultimate SD upscalers though.
J1B - does v2 require different prompting? it seems a little tricky to get the same quality on the same prompt vs v1. not necessarily less good quality per se but quite different and less detailed i think....
Not that I really know of, I do like to use long LLM generated prompts around 500 words.
I have taken out some more of the ZIT noise that is just faux detail really, like shown here: https://civitai.com/images/119135505
If you give me an example prompt I might be able to help.
But I guess if you like the look of My V1 ZIT more just keep using that :) it is quite subjective.
You could try adding in some of this film grain ZIT lora: https://civitai.com/models/693745?modelVersionId=2619388
It is not quite the same thing, but I like it:https://civitai.com/images/118463102
If you are looking for Photo realistic look , look at the prompt on this image with a tonne of keywords : https://civitai.com/images/119249678
"A photorealistic, ultra detailed, humorous scene on a bustling Dutch street market. A small young tabby cat anthropomorphized, running on two legs while tightly hugging a large, shiny silver fish. The cat has a determined, dramatic facial expression with wide eyes and an open mouth, as if mid-shout. Behind the cat, a shocked fish vendor in an apron is chasing after it, yelling. Fresh fishes are laid out on a market stall to the right, displayed on ice. The background features open stalls, market signs, and a few bystanders reacting in surprise. Dynamic lighting, rich details, cinematic composition, freeze-frame action shot, aspect ratio: 9:16, but don't give it a yellow/orange/brown hue. 8k uhd natural lighting, raw, rich, intricate details, key visual, atmospheric lighting, 35mm photograph, film, bokeh, professional, 4k, highly detailed, cinematic, colorful background, 8k, dramatic lighting, highly detailed, hyper realistic, intricate, intricate sharp details, fighting."
I have done some more testing of v2 vs v1 and yeah it is generally less detailed and photo real (apart from the backgrounds they are often more detailed)
@J1B thanks for that testing - i've gone through it a bit more and I think it's still very very good but just need to add a few more subject detail prompts i think vs v1.0. it's excellent though so muchas gracias for the checkpoint all the same.
imho: V-2 has better detail, V-1 has hotter women with same face, so V2 is not just better - it's different. But anyway, anything with the JIB prefix is the best generator of beautiful women :))
Yes Jib Mix ZIT V2 is probably a bigger step away from photo real look than I had indented, but moving any distance away from ZIT seems to lose some of that realism, maybe now we have ZIB training I can move the poses and faces away from ZIT without losing the photo realism, I just started on V2 before ZIB was released so I thought I might as well release it. I'm going to test which loras might bring back some photo realism and publish my results.
@J1B What I always see as a problem is that comparing the realism of generated images is very difficult. Where would you get the information from? If you look at "real" pictures taken with 10 very good digital cameras, you'll find 10 different levels of realism. Even if you take pictures with an analog camera, there are differences due to the lenses and the film carrier material. And when the super photo model comes out of the mask, she wears so much make-up that you can no longer see the pores of her skin, on purpose. Or the Instagram models who run 20 filters over the pictures then all the realism is gone. So i think... your models make very good Realism :)))
@Gorean "you'll find 10 different levels of realism." - true.
Will you make this available for generating images on civit?
I tried to, but Civitai do not support custom ZIT checkpoints (or Qwen-Image) yet. I think they do not have enough GPUs to power it.