NOTE: Civit has made the download UI confusing. For the v4.0 model, it contains four versions:
BF16 UNET (11GB)
FP8 UNET (6GB)
Q6 GGUF UNET (5GB)
FP8 Checkpoint (16GB)
I uploaded them very descriptively but Civit has stripped away the descriptions so it's now confusing to see what you're downloading. The FP8 checkpoint is an all-in-one version that contains the VAE and Text Encoder. For the UNETs you need to provide those extra models.
I've used 100s of curated images to help add some better color, contrast and detail to the Z-Image Turbo model. It's not a huge difference yet, but it is noticeably better in side by side tests. I'm starting out a bit conservative because I don't want to mess up any of the great compositional features that Z-Image offers.
Use the workflow embedded in the showcase images for the model version to get the best results. Check the About this version section for links to the workflow and recommended settings. Z-Image currently mangles nipples and genitalia so my workflow uses SDXL detailers to refine just those aspects of the image.
P.S. I would appreciate any constructive feedback on how to make this model better.
Description
v1.1 has been further trained with another 100 or so images from a curated dataset.
As with prior versions, I strongly suggest first trying the model with this workflow.
FAQ
Comments (13)
nice colors! seems to be working well with my character LoRAs in ComfyUI basic setup. Thanks for the model!
does it suffer from same face?
Only time will tell but, so far in my testing, I would say it has a bit more facial diversity than the original ZIT model.
1.0 is still the best at the moment. This is my personal opinion.
That's a perfectly valid opinion. Maybe I'll do a v2.0 Alt version that's a branch off of v1.0. Are you able to say what qualities of v1.0 that you prefer? (so I can try to retain those as much as possible)
It renders skin more realistically and works well with fingers, both hands and toes. Faces look very lifelike, in my opinion. When comparing versions 1.0 and 1.1, I noticed more plastic-like skin and less detail in 1.1. In 2.0, there's more contrast and still less detail, although the faces are more stylish. I also prefer the lighting in 1.0—it's more accurate, without distortion. I don't speak English well, so I use a translator.
@monkmb Thanks for all the feedback. I appreciate it!
For some reason, character LORA doesn't work with Zimage custom checkpoints.
What do you mean? I've been creating and using character loras with my checkpoint.
@GBRX the images are deteriorating
@zagzagem685 Try turning the lora strength down to under 0.50. With Z-Image I've noticed that higher lora strength causes problems. I have the best luck with 0.25-0.5.
Another impressive model. Good color and saturation balance. I think the base model can be a bit to grainy and muted at times, and this is just the right amount of "pop". So i totaly get the naming of the model.
Thanks, that's how I feel about the base ZIT as well. It's just needs a little something extra.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.