It seems like AI image generation is close, but still slightly "off" as you say. To me it remains to be seen whether that's actually a huge achievement, or whether there's more work to be done to achieve that last 10% than has been done so far. The poster you're replying to used gaming as an example, and you only have to look there to see how graphical improvements in the last 10 years have diminishing returns versus earlier improvements.

I could be wrong but this seems like an area where our progress will look like an S-curve. Getting that last 10% could be the real achievement.