A default workflow for Z-image, modified to do Image2Image and for gguf model.
Denoise is the way.
From 0.4 to 0.74. Strictly depends on source image.
Low value is good for upscaling image.
Higher value is like t2i.
Description
Initial workflow
FAQ
Comments (10)
where do i get the model needed in the "Unet Loader (GGUF)" node? ty
You can find the quantized versions here:
https://huggingface.co/jayn7/Z-Image-Turbo-GGUF/tree/main
If you’re able to run the default model, you should be able to swap it out by using a Load Diffusion Model node.
@GrumpyNSFW sorry, i have to ask you about other two files. Cant find them with gogle and hugginnface. Qwen3-4b-UD-Q4-K-XL.gguf and zimage_vae.safetensors. Is this ok https://huggingface.co/unsloth/Qwen3-4B-GGUF/tree/main ? And exact same vae. file unreached by me...
Hi, I am trying your workflow in ComfyUI, I downloaded all your models but for some reason I am getting this error: "Unexpected text model architecture type in GGUF file: 'qwen3'" in CLIPLoaderGGUF, any ideas why this occurs?
You need a qwen3 clip. https://huggingface.co/unsloth/Qwen3-4B-GGUF/tree/main
update gguf loader in comfy manager.
@shnack I updated my GGUF Loader in my COMFYUI and worked! Thank you!
just writing here to say thank you OP for asking this and thanks to @shnack for the solution!!!
Getting error:
CLIPTextEncode
gather(): Expected dtype int64 for index
Any suggestions? Comfy-desktop running latest and i've updated the only custom node in the workflow as well.
you have to update comfyui, running the .bat file in update folder.