>>106977879
I'll *talk* slowly
Q8 is less accurate and so retains less details
nunchuku is more accurate and retains more details
nunchuku contains many weights that, when changed, caused visual changes, and so were instead kept at fp16, other weights that have little or no quality impact are kept at fp4, this shift causes small variances in generations, but maintains quality

Did that work for your little reptile brain?