how to use the fp16 safetensors within comfyui-qwenvl prompt enhancer?
the model details are captured in the hf_models.json file:
"Qwen3-4b-Z-Image-Engineer-V4": {
"repo_id": "BennyDaBall/Qwen3-4b-Z-Image-Engineer-V4",
"default": false,
"quantized": false
},
the 'auto-download' from the node doesn't work:
ValueError: Couldn't instantiate the backend tokenizer from one of:
(1) a tokenizers library serialization file,
(2) a slow tokenizer instance to convert or
(3) an equivalent slow tokenizer class to instantiate and convert.
You need to have sentencepiece or tiktoken installed to convert a slow tokenizer to a fast one.
if i use the manual way:
hf download BennyDaBall/Qwen3-4b-Z-Image-Engineer-V4
It gets downloaded under C:\Users%USERNAME%.cache\huggingface\hub\models--BennyDaBall--Qwen3-4b-Z-Image-Engineer-V4
but its still not usable from within the comfy node
The .safetensors in this repo are not intended to be used in ComfyUI. They are sharded safetensors for people who want to train the model or run inference elsewhere. I only added them so people could tweak the model and quantize it the way they want.
Use City96 GGUF clip node and the GGUF file version of Z-Image Engineer from this repo (saved in your text_encoders folder)!
Gotcha, i was trying to not use GGUF, and stick with FP16.
What's curious is that, the official ComfyUI-QwenVL repo, has your model included in the hf_models.json file.
ref: https://github.com/1038lab/ComfyUI-QwenVL/blob/main/hf_models.json
Gotcha, i was trying to not use GGUF, and stick with FP16.
What's curious is that, the official ComfyUI-QwenVL repo, has your model included in the hf_models.json file.
ref: https://github.com/1038lab/ComfyUI-QwenVL/blob/main/hf_models.json
GGUF is available in FP16 - no quality loss! Ahhh, they made a mistake lol... that repo doesn't contain a .safetensors model. They should be pointing to https://huggingface.co/BennyDaBall/Qwen3-4b-Z-Image-Engineer-V2.5/tree/main/safetensors_format for V2.5 or https://huggingface.co/BennyDaBall/Qwen3-4b-Z-Image-Engineer-V4/tree/main/safetensors for V4. You should be able to manually download the files and add them to the appropriate folder per the ComfyUI-QwenVL instructions.