In this video, Philip Dyer compares the prompted output of three popular AI image generators: Dall-E Mini, Midjourney, and Dall-E 2. Dyer's interest is photonormative output, but it goes places and you get the picture.
I pit these 3 AI image generators against each other to see which reigns supreme. Spoiler alert: DALL-E 2 wins.
I think Dall-E Mini's crappy Francis Baconator style is what's actually popular (not least because it is accessible) and not to be dismissed because of its lack of realism. To quote Brian Eno: "Whatever you now find weird, ugly, uncomfortable and nasty about a new medium will surely become its signature."
Midjourney is for generating cursed concept art.
Dyer is right in that Dall-E 2 is where it's at if you want superficially convincing photos. I could imagine using to expedite digital art commissions in specific styles or genres, too. Its output is good enough to paint over. It (or whatever like it gets released to the general public) will soon replace "photobashing" as the thing amateurs and wannabes think is beneath them, but pros do all the time.
P.S. Realism is a red herring, a gamer-generation idée fixe and a TV-generation paranoia, which forgets how easy it is to fake realistic output without help from AIs. Consensus reality is dissolved by hand-crafted lies, network effects and the economic implosion and consolidation of news media, not these AI image- and text- generators.