Most AI Detectors are garbage
Also
View attachment 4863431
Well, that seems to settle/answer the question. From his (or her) own statement, can be assumed everything is "AI"-generated, just it is the quantity of editing after that is variable.
But quantity of editing is relative, can be very high (so, still considered anyway an artist doing stuff, even if helped) or not.
Since the majority of the systems are "opaque"/"black boxes", sometime it is sufficient to change two pixels (there is even a "one pixel attack") to make a system recognise an image as something completely different from what clearly is for us humans, and something similar apply to systems trying to recognise if it is "AI"-generated stuff, it may be enough to change a few pixels to fool the system in saying the "AI" probability is low, even if it is "AI"-generated.
Aside from the infamous "7 finger hands"

, which has improved (in these pictures, they all five fingers or are not visible), often a clue is coloring and style - despite the name, "generative AI" do not generate anything from scratch.
They are machine-learning systems (not "AI" proper, now often labelled a "general AI" or "generalist AI" due to the overabuse of the "AI" label, which does not exist, at least not yet) that learnt patterns from images given to them, take elements from those patterns and try to produce a result that seems to correspond to what the text describing the request indicate.
Text that itself is actually analysed by a machine learning-based system that has learnt from text given as input combination of words that are expected to go together and what can be an output that satisfy that word combination.