Flats doing my head in

[Datalord]

I need some help pinpointing this problem. I know my WO71 scope has some issues in the imaging path, but I've been able to take some decent pictures with it in NB. However, lately I had fun with M31 in LRGB and as I started processing the lum my flats went nuts on me.

I have used the same set of images (bias, dark, flat and light) to produce the two different stacks below. One is PI batch stacked, the other is DSS. I did an automatic background extraction on both and STF them, and I STF'ed the master flat they produce as well beside them.

DSS:

PI:

What's doing my head in is the difference in stacked pictures and how it looks. I can work with the DSS version with some masking trickery, but the PI stack is total nonsense to me. If anyone has some golden nuggets of advice, I'm all ears.

For comparison, here is a PI stacked Ha in progress of Elephant trunk. None of this madness I'm seeing in the other stacks. All images and stacks were made with the same setup, no differences in moving camera around or anything. Different flats, obviously.:

 

[Shibby]

Both flats look very similar to me, so surely something else is going on with the second image? You say you used DBE... The dark area around the galaxy in the 1st image looks typical of DBE when it gets confused and includes the object in its measurements of the background. Have you tried without? The second image, not a clue what's going on there.

[Datalord]

15 minutes ago, Shibby said:

Both flats look very similar to me, so surely something else is going on with the second image? You say you used DBE... The dark area around the galaxy in the 1st image looks typical of DBE when it gets confused and includes the object in its measurements of the background. Have you tried without? The second image, not a clue what's going on there.

Right, here is the DSS without DBE, with STF. By no means a perfect starting point.:

[Datalord]

With massive amounts of tinkering with DBE, I got the below result. Probably better than the automatic, but this will definitely not be an easy process.

[Datalord]

Well, please ignore an idiot. I integrated a set of images in PI using old, wrong flats. The proper PI image looks pretty much like the DSS stack. It still is bad, but not deformed. ?‍♂️

[Datalord]

Well, I thought I could handle this remaining stuff, but this being my first RGB image, I am pondering whether I just never ever do RGB again or if my current skies are too polluted for it. This is the result of Lum + RGB:

Send help...