It's not that these images are perfect - it's that they're close enough.
The "problem" is that these images look amazing with a minimal touchup - something which would happen anyway to a real photo.
An extra hour to two fixing some AI artifacts (the ol' droop-eye and derp-hand) is a LOT cheaper than getting actual people out to an actual location and taking an actual photo.
I think Zuck is right about this, insofar as the comments in the article are concerned.
OpenAI have done a brilliant job of selling the dream - but there will not be one "god model," - there will be many specialised, smaller models.
You can already see it going that way with new hardware shipping with NPU's. These workloads are expensive to run and shipping them to your device is a top priority.
I get what your saying, in that open source projects normally have a licence that applies to how itās used - but this has always been open to abuse.
Nothing has ever stopped things like this happening - see how industry has taken advantage of open source for decades (often productising things as their own in the process).
The eyes see what they want to see š¤·āāļø