mistral_3_small_flux2_fp4_mixed

clip model fp4 for FLUX
fits into 16GB VRAM very fast with new prompt handling

"10times faster if you change the promt" only in the this case clip model working.
you can easy replace fp8

all About this license is Apache license 2.0 from mistral Ai
and convert by

origignal source: https://huggingface.co/Comfy-Org