GGUF experiment.
I'm working on an experiment involving text encoding (for Flux.2 Klein 9B) and llama-cpp-python using this Qwen3-VL 8B.
I've tried using some gguf quantized directly from qwen/qwen3-vl, like for example Qwen/Qwen3-VL-8B-Instruct-GGUF, and even though it works, it produces a really bad quality image.
It also produces a bad quality image with a standard CLIPLoader (GGUF) node. So that's not just my experiment (which probably needs some more work anyway).
I'm suspecting that it might be the model itself that's different. So have you done anything particular to make qwen3-vl work with Klein 9B ? If so, how should I proceed to obtain a GGUF of qwn3-vl 8b that will work properly with Klein 9B ? Any idea ?
While testing my stuff, I realized I must have messed up a little bit, as qwen3-vl-8b (any variant) produces bad quality images (with my experiment or with regular clip loaders).
Almost a big waste of my time, almost because when making the same experiment with a qwen3_8b model, I managed to get both a good picture and generated text within 5sec, instead of the long generation you usually get with the comfyui Generate Text node.
half-big sad. lol
Sorry for wasting your time.