Join The Tinkerer on Whop. Membership gets you early releases, private tools and a bunch of extra stuff.
👉 Join on Whop
💬 Join the community for support, free tools and early news on Discord
A clean and practical CyberRealistic Z-Image Turbo workflow for ComfyUI, built for strong results without making things unnecessarily complicated. The workflow is designed to stay user-friendly while still offering useful extras such as upscale options, cleaner organization, and improved saved-image metadata.
CyberRealistic Z-Image Turbo - ComfyUI Workflow V8
This workflow is built for power users who want fast access to the parts they actually use: prompt editing, prompt engineering, previewing, comparing, and saving results.
The Prompt and Save Image sections sit close together on purpose. You can add or adjust a prompt, generate, check the result, and save the final output without scrolling across the whole canvas. Everything else stays fully accessible, including detailers, tiled upscaling, SeedVR2 upscaling, compare previews, LoRA loading, and advanced settings.
What's new in V8
SAM3 detailers. Face, Eye, and Hand detailing no longer rely on YOLO bbox models. Detection now runs through SAM3 with a plain text prompt, feeding MaskToSEGS and DetailerForEach. More accurate on unusual poses, crops, and stylized subjects.
Subgraphs. Each detailer is packed into a subgraph with every relevant setting exposed on the node face. Threshold, crop factor, drop size, guide size, denoise, feather, and wildcard are all editable without opening anything.
Number of characters. One integer drives how many detections each detailer expects. Faces scale with the count, eyes and hands scale with double the count. Set it to the number of people in your image and the detailers follow.
Additional Detail module. A separate optional module with dedicated detailers, including its own group toggle panel so you can enable only what you need.
DisTorch2 model loading. The UNET loads through ComfyUI-MultiGPU with a Virtual VRAM control. Set it to 0 on 16GB or more, 4 to 6 on 12GB. Adjust to your own machine.
NegPiP prompt encoding. Negative weights inside the positive prompt, encoded through a dedicated Z-Image NegPiP node.
Sage Attention and Torch settings patches for faster sampling.
Module toggles 1 to 6 in one panel: Face Detailer, Eye Detailer, Hand Detailer, Additional Detail, Upscaler, SeedVR2.
Detailers and SeedVR2 are disabled by default. The upscaler is enabled. Save Image sits at the end of the chain and follows whatever modules you turn on.
Requirements
A recent ComfyUI build with subgraph support and the core SAM3_Detect node
CyberRealistic Z-Image Turbo v8.0 (this workflow points at the int8 build, rename or repoint the loader if you use another variant)
Qwen3 4B text encoder and the Z-Image VAE
SAM3 checkpoint (sam3.1_multiplex_fp16) for the detailers
4x_NickelbackFS upscale model
SeedVR2 DiT and VAE only if you use module 6
Nodes / Custom Nodes Used
rgthree-comfy
https://github.com/rgthree/rgthree-comfyCyberdelia Z-Engineer
https://github.com/cyberdeliaAI/comfyui-cyberdelia-z-engineerCyberdelia Metadata / Save Image With Metadata
https://github.com/cyberdeliaAI/revived_comfyui_image_metadata_extensionCyberKrea Sampler (empty latent and cyberzit schedulers)
https://github.com/cyberdeliaAI/ComfyUI-CyberKrea-SamplerNegPiP for Z-Image
https://github.com/cyberdeliaAI/comfyui-negpip-zimageComfyUI Impact Pack
https://github.com/ltdrdata/ComfyUI-Impact-PackComfyUI Impact Subpack
https://github.com/ltdrdata/ComfyUI-Impact-SubpackComfyUI Ultimate SD Upscale
https://github.com/ssitu/ComfyUI_UltimateSDUpscaleComfyUI KJNodes
https://github.com/kijai/ComfyUI-KJNodesComfyUI-MultiGPU
https://github.com/pollockjj/ComfyUI-MultiGPUComfyUI-Easy-Use
https://github.com/yolain/ComfyUI-Easy-UseComfyUI_essentials
https://github.com/cubiq/ComfyUI_essentialsComfyUI Comfyroll Studio
https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodesSeedVR2 Video Upscaler
https://github.com/numz/ComfyUI-SeedVR2_VideoUpscalercomfyui-vrgamedevgirl
https://github.com/vrgamegirl19/comfyui-vrgamedevgirlAnima / Z-Image Resolution Node
https://github.com/cyberdeliaAI/ComfyUI-anima-Resolutions
Note: RES4LYF is no longer required in V8.
Future updates and beta versions are available via the public Whop page:
https://whop.com/cyberdelia-ai-lab/comfy-ui-workflows-zit-C80otuTsikeE8d/app/
Description
Future updates and beta versions are available via the public Whop page:
https://whop.com/cyberdelia-ai-lab/comfy-ui-workflows-zit-C80otuTsikeE8d/app/
FAQ
Comments (18)
This is great! I'd like one with random image resolution! Do you have something equivalent for all pony, sdxl, illustrious models?
i was looking at this right now, i mean with the llm..
@LerestesunirIA Ah got it. Yeah maybe in the future. I use LLM for prompting with Pony/Illustrious.
Great workflow and model. Thanks for providing these! I'm having an issue, though.
If I enable the Skin Enhancer node, I get a PyTorch error:
pickle.UnpicklingError: Weights only load failed. In PyTorch 2.6, we changed the default value of the weightsonly argument in torch.load from False to True. Re-running torch.load with weights_only set to False will likely succeed, but it can result in arbitrary code execution. Do it only if you got the file from a trusted source. Please file an issue with the following so that we can make weights_only=True compatible with your use case: WeightsUnpickler error: Unsupported operand 60 Check the documentation of torch.load to learn more about types accepted by default with weights_only https://pytorch.org/docs/stable/generated/torch.load.html.
Looking at the traceback, it seems to be coming from ComfyUI/comfy/utils.py. I tried changing weights_only=True to False, but then I get a different PyTorch error.
I've pulled the latest ComfyUI repo and used pip to rebuild all the dependencies. No change in behavior. Any suggestions other than just leaving Skin Enhancer off?
This error ("Unsupported operand 60") almost always points to a corrupted or incomplete download rather than a real PyTorch/ComfyUI bug especially since the model loads fine for others and you're already on latest ComfyUI. The fact that flipping weights_only=False produces a different error supports that: once the safe-load check is bypassed, the unpickler trips on the actual damage in the file.
Try this:
Delete 1xSkinContrast-SuperUltraCompact.pth and re-download it fresh from OpenModelDB. The file is small, so a partial download isn't obvious from the size alone.
Just to confirm on my end the workflow runs fine here with the Skin Enhancer enabled, so the issue looks specific to your setup (most likely a corrupted download of the upscale model, see my other reply).
That said, I rarely use the Skin Enhancer myself, so if the fix turns out to be a hassle, leaving it off isn't a real loss.
@Cyberdelia Thanks so much for the quick replies! I replaced the 1xSkinContrast...pth file and it works now. First impression is it doesn't seem to do much for the prompt/output. But, it's nice to know the error is gone if I do want to use it.
Again, greatly appreciate your workflow and model...and now customer support. ;)
Hello, awesome workflow! 🤌 Any idea how to resolve the "AttributeError: 'coroutine' object has no attribute 'outputs'" on the SaveImageWithMetaData node?
I think this is a small bug in some situations. I will push a fix.
It's updated now. So it will not give this error on your ComfyUI version.
You can update via ComfyUI-Manager or 'git pull'
https://github.com/cyberdeliaAI/comfyui-cyberdelia-metadata
Thanks for the reply! I've updated ComfyUI, and there are no updates for the custom nodes. Still getting the error, but I think I've isolated the error to the SaveImageWithMetaData node itself. I tried running a workflow I got many months ago, and it is failing for the same error.
UPDATE: I've figured out what happened - I still had the SaveImageWithMetaData node from edelvarden, who is no longer updating the node, apparently. I also found out that I did not have your version of this node (ID 3440 in the ComfyUI Manager) completely by accident while looking for the edelvarden version to disable it after getting some help in the Civitai Discord server. I installed your version, and once I disabled the edelvarden version, your version became the default and I am no longer getting that error.
It's very awesome, but how do you guys keep LM studio from complaining about NSFW stuff? I am using Gwen3.5 9B and it is like talking to a nun or something
Use a unrestricted model (Heretic version).
UGI leaderboard, sort by willingness. I was never here.
you could also 2 stage it, use cyberrealistic to generate base image, re-encode it for z-image, then z-image text encoder only see's 'enhance image, repair image... etc'. it's used as a refiner.
This is too fun, ty buddy. <3