This is a GGUF versions of Z-Image-Base – The non-distilled foundation model.
Guidance scale: 3.0 – 5.0
Inference steps: 28 – 50
Sampler: res_multistep
Model (6.73 GB) - GGUF Q8_0
Model (5.68 GB) - GGUF Q6_K
Model (4.72 GB) - GGUF Q4_K_M
Original HF Repo: https://huggingface.co/Tongyi-MAI/Z-Image
Description
FAQ
Comments (5)
Why am I getting this error? I have the correct workflow. : UnetLoaderGGUF / LoaderGGUF Error(s) in loading state_dict for NextDiT: size mismatch for x_pad_token: copying a param with shape torch.Size([3840]) from checkpoint, the shape in current model is torch.Size([1, 3840]). size mismatch for cap_pad_token: copying a param with shape torch.Size([3840]) from checkpoint, the shape in current model is torch.Size([1, 3840]).
Try updating the ComfyUI-GGUF loader nodes or use a different GGUF loader implementation.
I use the city96/ComfyUI-GGUF (https://github.com/city96/ComfyUI-GGUF).
Your error looks like an incompatibility between the GGUF loader and the specific GGUF file format - the loader expects different tensor shapes than are in the checkpoint. In other words, the GGUF file was likely built in a slightly different format that your current loader node version does not support yet.
@Mr_Einhorn I'll try that, and I'll let you know if it works, THANKS
I'm a little confused about the description. does the Q8 (the 6.8GB) is a gguf version of FP8, or it is gguf version of Zimagebase BF16?
All GGUF versions produced from Zimage base BF16. Unfortunately, when uploading models to Civitai, you’re required to specify different quantizations for the uploaded models; otherwise, it caused an error. Now we have a leaked FP32 version, is it worth making that too?