If we take the AI industry’s reports in good faith, not counting the training, the computer the prompting was done wasn’t used for other things afterwards, the AI one-shotted the image and the user didn’t ask for revisions (which “anti-slop” AI boosters are against), etc.