Interesting to see a bit of the process, and I think one of the issues might be the compression-induced "vaseline filter" on the image in the original post? Your post has a much less compressed version that's much sharper and thus more *obviously* not AI (My initial verdict was "probably human, low confidence" due to the compression making it hard to *confidently* determine "human" until your post)
(But I can *see* where the original accusation was coming from, even if I would never have made it myself: with the heavy compression in the original post it's *much* harder to determine if possible "tells" are actually *tells*, or just Compression...)