Hugging Face's logo

unsloth
/
FLUX.2-klein-9B-GGUF

Image-to-Image
GGUF
Diffusion Single File
English
ggml
image-generation
unsloth
image-editing
flux
Model card Files
xet
 Community
9

Which text encoder matches flux-2-klein-9b-Q4_K_M.gguf in ComfyUI-GGUF?

#6
by aperturezhang - opened
Discussion
aperturezhang

Hi, I’m trying to use flux-2-klein-9b-Q4_K_M.gguf from this repo in ComfyUI on macOS with ComfyUI-GGUF.

My environment:

  • ComfyUI desktop 0.8.28
  • ComfyUI core 0.18.5
  • torch 2.12.0.dev on Apple Silicon (M2 Max, 64GB)
  • ComfyUI-GGUF installed

I used:

  • UNet loader: UnetLoaderGGUF
  • UNet: flux-2-klein-9b-Q4_K_M.gguf
  • VAE: flux2-vae.safetensors

I could not find the exact matching text encoder for this GGUF model.

I tried:

  1. qwen_3_8b_fp4mixed.safetensors
  2. qwen_3_8b_fp8mixed.safetensors

Both fail with text encoder shape mismatch errors in ComfyUI, for example:

RuntimeError: mat1 and mat2 shapes cannot be multiplied (512x4096 and 2048x1024)

Could you please tell me the exact text encoder file that matches flux-2-klein-9b-Q4_K_M.gguf in ComfyUI-GGUF?
If possible, please share the exact model filename and loader combination.
Soup091

try this : https://huggingface.co/Qwen/Qwen3-8B-GGUF/tree/main
BingoBird

I got results with this as well: Qwen3-8B-abliterated-josiefied-goekdeniz-v1-IQ3_XS.gguf

I'm not sure whether 9/10 gens creating physical abominations is normal here. I can only run the Q4 GGUF in stable-diffusion.cpp sd-cli.

Sign up or log in to comment