is there a text encoder issue for this model right now?
i cannot get the mistral small 3 fp8 text encoder to work at all with any of the ggufs or the full model. im not sure why because it works with the flux 2 dev gguf but not this.
fails right at the step count on all of them
mat1 and mat2 shape error.
its the gguf node not being pr updated to run the model. guessing i just have to wait in this case.
These models do not use mistral. the 4b klein uses https://huggingface.co/unsloth/Qwen3-4B-GGUF and 9b uses https://huggingface.co/unsloth/Qwen3-8B-GGUF
4b is fine!
8b - gives an error!
SamplerCustomAdvanced mat1 and mat2 shapes cannot be multiplied (512x12288 and 7680x3072)
4b is fine!
8b - gives an error!
SamplerCustomAdvanced mat1 and mat2 shapes cannot be multiplied (512x12288 and 7680x3072)
flux2 klein 4b with qwen3 4b, and flux2 klein 9b with qwen3 8b, should work.
4b is fine!
8b - gives an error!
SamplerCustomAdvanced mat1 and mat2 shapes cannot be multiplied (512x12288 and 7680x3072)flux2 klein 4b with qwen3 4b, and flux2 klein 9b with qwen3 8b, should work.
Flux 2 klein 4B Q8_0 + Qwen3 4B Q8_0 here - same error.
I guess GGUF users have to wait.
You may need to update comfyui and run based on the latest from the repo. It doesn't work if you don't update.
This image was created with the exact combo (Flux 2 klein 4B Q8_0 + Qwen3 4B Q8_0).
1)Update comfyui.
2) "change the preview method from auto to none". (this will fix the errors). if you have animated previews you may also check VHS settings.
Bigger model works with 8b text enc, smaller model works with 4b text enc. However the anatomy knowledge sucks for flux models again vs qwen image and z image models. You can use the gguf models in combination with safetensors(text encoder) as well.