❤️ Please Consider Supporting My Work
I am passionate about making open tools and uncensored models for the community. If you value this work and want to support independent development, please consider buying me a coffee to keep the momentum going!
👉 https://ko-fi.com/lorian0x7
Fix Anatomy, Improve Skin Texture & Enable Output Variety for Z-Image
We all know that Z-Image is a game-changer, but even the best models have their weak spots. Currently, it still struggles with proper anatomy for "spicy" details. This workflow is designed to solve that problem.
It acts as a second-pass refiner (inpainting) that targets the body to correct anatomy and apply a layer of hyper-realism while preserving the composition of your original generation.
✨ NEW UPDATE: Wildcards
I have added Wildcards to the workflow! These wildcards are specifically optimized for Z-Image to significantly improve the variability of each seed, giving you distinct and unique images with every generation.
Note: Version 1.0 (without the wildcard body refiner) is still available for download if you prefer the simpler setup.
The Core Model: Uncontaminated Realistic
To achieve the results seen in the preview images, this workflow relies on Uncontaminated Realistic.
This is a specialized model based on the Illustrious architecture. I created it to bridge the gap between the insane flexibility of anime models (poses, tentacles, niche tags) and photorealistic rendering.
⚠️ Model Availability: Uncontaminated Realistic is currently an exclusive model available only to my supporters on Ko-fi.
Alternative Models: While this workflow is optimized for Uncontaminated, it is fully functional with other Illustrious-based realistic models. However, please note that results may vary.
📂 Installation & Requirements
To use this workflow, you need to download the required detection models and place the new wildcards in the correct folders.
1. Detection Models
You need the YOLOv8 model and the Body Detection model.
YOLOv8: Download Here
Female Body Detection: Download Here
Place them in the following folders:
YOLOv8:
ComfyUI/models/ultralytics/bbox/Body Detection:
ComfyUI/models/ultralytics/segm/
(Make sure to select the correct model in the Ultralytics provider node inside ComfyUI).
2. Wildcards (New!)
The included wildcards are essential for the new variability features.
Place them here:
ComfyUI/custom_nodes/comfyui-impact-pack/custom_wildcards
(You have to restart comfyui to refresh the wildcards)
🔗 Support & Download
To get Uncontaminated Realistic and unlock the full potential of this workflow, visit my Ko-fi page:
https://ko-fi.com/lorian0x7
I’d love to see your results in the gallery!
Description
FAQ
Comments (13)
Does it seem like both 'zimg' and your 'Uncontaminated Realistic' need to be loaded simultaneously? How much memory and VRAM would be required to run them?
that's correct, I'm running it with 24gb vram but you can run it with less if you have free ram the model remains cached there and fast loaded in the vram when it's needed.
I think if can run z-image and you have free ram, you should be fine
@Lorian Alright, thank you for your reply. Can it be used together with the GGUF model of zimg?
@Black130516 yes, no problem at all, just swap the model loader node with the gguf one
@Lorian Nice, so how do I get the 'Uncontaminated Realistic' you mentioned? Do I just need to sponsor you?
@Lorian Hey, when I use this workflow, I always encounter the error at the final step: "'NO_SEGM_DETECTOR' object has no attribute 'detect'". I've tried versions a, b, and c, but none of them work.
Found the issue—segm should be placed in the segm folder, not the bbox folder. It's resolved now.
@Black130516 Ah thank you for spotting this issue, I'll update the description. And yes, for uncontaminated realistic just support my work on kofi.
@Lorian I've already sponsored you and sent a private message. :)
@Lorian What's your kofi account? https://ko-fi.com/lorian doesn't seem like it.
@tubbzo it's this
https://ko-fi.com/lorian0x7