[dragonwriter]

> Would they have noticed them at the quick glance that this image invites?

Probably; I mean, it took me more than a quick glance to even figure out that the thing that looked like a discontinuous strip of printed cash register receipts with red stains across the outside of his sweater was probably supposed to be bandages, even with the priming effect of reading the text. And the document wrapped partly around his arm is quite jarring.

But, ultimately, may main point isn’t that AI images are categorically unconvincing—I’ve generated more convincing AI images. But AI barely matters; the same training data that enables AI image generation contains thousands of images that can be used, without modification, with a false caption to the same effect — and not just in theory, but this is actively done all the time. While AI may increase the risk of convincing fake imagery of specific individuals (though that, too, has been common before AI image generation), the kind of generic propaganda highlighted here is both simple and a hundreds-of-times-a-day thing with more convincing imagery without AI.