i can reproduce this using a custom build on a QT Py that adds the native _pixelbuf. it does not occur using brightness=0 in the constructor:

import board
import neopixel
np = neopixel.NeoPixel(board.NEOPIXEL, 1, brightness=0)
# no change visible yet
np[0] = (100,0,0)  
# now turned off
np.brightness = 0
# back to full on

looking in shared-bindings/_pixelbuf/PixelBuf.c, the constructor calls mp_obj_get_float(), while the setter uses mp_obj_float_get(). this skips ..._get_float()'s extra error-checking that seems to catch this case for the constructor. i can't follow the bitmath in mp_obj_float_get() to prove the fault, but testing shows that the setter takes the brightness > 1 fork after it's called.

changing the setter to use mp_obj_get_float() instead of mp_obj_float_get() does seem to fix the error, but it will add some cycles. changing the bitmath might be another option but i don't have any suggestions as to how.

@brrzap good catch, wanna submit a PR for that setter change? its ok if it adds some cycles - better to have it right :)