well, you can try with LN again, but the trick is twofold - one is to make sure you are using your absolute best subexposure for the reference, meaning the most gradient-free sub, also taking care that it looks like it was flattened properly. if necessary you can do DBE to that reference sub but you have to be very careful - under LN all the subs will take on the character of the reference and so any artifacts you introduce into the reference will manifest in the other subs.
the other trick is to let the LN tool create a normalized frame that you can inspect. if there are artifacts you can try adjusting the scale parameter. i think RickS had mentioned in another thread that larger scales (like 256) were working well for him, but this could be OTA/Sensor dependent. i noticed in one of your images there were large-scale dark artifacts around some of the stars, which looked to me like LN artifacts (which is why i suggested trying with it turned off...)
note also that the first frame in the list of files in ImageIntegration is the reference frame for the standard normalization algorithms, so it probably pays to have the highest quality sub first in the list when not using LN.
rob