I got a ‘human made’ image wrong because areas of detail became a smeary pattern. The problem was very likely caused by computer-made compression artifacts, and

I got a ‘human made’ image wrong because areas of detail became a smeary pattern. The problem was very likely caused by computer-made compression artifacts, and I wouldn’t be surprised if similar generational AI becomes part of immediate post processing of cell phone snapshots. It makes me wonder, are models being unintentionally trained to produce jpg artifacts? How much better would a generator be if it had been trained on, say, pngs?