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
Better NSFW details, although a little light SDXL refinement never hurts. You can use the workflow from this image, https://civitai.red/images/130290388, and the Ultralytics bbox files from https://civitai.red/models/2172100/gonzalomo-z-image-refiner-workflow?modelVersionId=2543959. Personally, I like to use my MoP v7.1 model to do the refining but you can use any photorealistic SDXL checkpoint.
Thanks to @nobodybutmeowie, @purplelady and @Disruptor_art_warehouse for the prompt inspiration!
Normally I have a small set of suggested settings but I'm having a hard time finding any samplers and schedulers that won't work with this model. Maybe it's just a z-image thing but I would encourage you to try out any set of settings you can think of and just see what happens.
Here are the settings that I'm using most frequently:
sampler: sa_solver, euler, dpmpp_2m, er_sde, res_2s
scheduler: beta
cfg: 1.0
steps: 5-12
There are four versions available for download:
BF16 Unet
FP8 Unet
Q6_K GGUF Unet
FP8 Checkpoint that also includes the UltraFlux VAE and qwen3_4b_fp8_scaled text encoder.
FAQ
Comments (23)
Yo Gonzo. You're recommending an AIO workflow for a diffusion model. That's why people are having the crashes.
Look at what I did on mine and throw a switch in there so you can use it for both.
Thanks, I removed the workflow link from the model details since I have a per-version workflow now in the About This Version section. The link in that section is correct.
V4, Very nice, my friend!! Nice faces and bodies, good lighting and contrast... Well done! Thanks! - Geno
Well done. The model responds well to age queries, executes the script correctly.
As far as I can tell, it handles the script better (because it responds better to age queries) than Qwen,
Great, thanks!
the workflows doesn't work. throws bunch of errors
Can you please be much more specific? Which model version and which workflow are you using? What are the errors?
It looks very nice. Pls could be possible, that you will prepare extra zip with workflow and additional required files? I'm fighting with comfyUI and not able to get it run. Comfy is a huge challenge. Or if it could be possible to zip complete portable comfy release (just without the main checkpoint), so there will be nothing missing...
Hey GBRX, thank you for sharing your models- 4.0 is currently my favorite one and it goes really well with my style of prompting. you inspire me to get better at this. Much appreciated!
The quality of your ZIT models is impressive. They are my favorite on civitai. Thanks for sharing them with us.
Thanks
Your v4 model is giving me this error in ForgeNeo: RuntimeError: The size of tensor a (2560) must match the size of tensor b (160) at non-singleton dimension 1
Which download version are you using? BF16, FP8 unet, Q6 GGUF or FP8 checkpoint?
@GBRX It was the fp8 pruned 16GB version. I will try the 11GB fp16 version next.
bf16, 11GB, seems to be working, thanks.
@olternaut Thanks, I've added some text to the model description to make the civit download UI less confusing.
Thanks
I've been trying other models and style loras and find myself coming back to this, by far my favorite. Are there any plans for flux 2 klein 9b?
Je l'ai testé, c'est excellent, le rendu est supérieur à Z-image Turbo... c'est devenu mon préféré, merci et très bonne journée