After I had finally managed to craft this beast, I felt so proud that I just had to share it:
/(<img([^\/]|\/(?!>))*)>/U
If I were to try and translate this into common language, it would perhaps go like this: after each occurrence of <img, skip all non-/ characters, and also all occurrences of / immediately followed by a > — up until the next occurrence of >. Unless you have a match at that point — a > without an immediately preceding / — move on to the next occurrence of <img.
And with each part of the expression in parenthesis, between quotation marks, following the corresponding translation: after each occurrence of <img, skip all (“*“) non-/ characters (“[^\/]“), and also (“|“) all occurrences of / immediately followed by a > (“\/(?!>)“) — up until the next occurrence of >. Unless you have a match at that point — a > without an immediately preceding / — move on to the next occurrence of <img.
The outermost parentheses (/(<img...)>/U) grab the contents of the unterminated <img> tag apart from the closing angle bracket, so that it’s easy to pair with a “ />” to properly terminate it. Other parentheses are for defining subpatterns.