So it looks like so far three keypoints make a big difference: 1, tilted ultra wide lens with foreground close and sharp and 2, no plants because immediately it looks fake (it looks fake even if you add plants in the foreground to me) and 3, Geocontrol.

Point 1 with the only exception of MoppelundWilli work because he doesn't use a tilted ultra wide yet  is images are ultra realistic.

Speaking of Geocontrol, is version 2 really necessary or version 1 will do? 129 euro is quite a bit of money for me now. I am quite confused about this software, since many links in this thread show jaw dropping work NOT done with the help of Geocontrol.