There’s an ongoing thread on Twitter where Sam takes suggestions from replies and shares the output. E.g. https://x.com/sama/status/1758200420344955288?s=46&t=VQo1eLU...
Even these may be cherry-picked though, he's only posted a few and I'm sure he's gotten thousands of requests already.
I guess he might be generating 50 for each response and posting the best, but that would seem deliberately disingenuous which hasn't been openai's style.
even the worst is still orders of magnitude better than anything else.
2nd is quite simple, but still suffers from your typical lighting issues that plague image gen. (shadows are significantly off)
3rd has magically appearing spoon and isn't that complex.
4th has a lot of prompt following issues
Some others feel quite off -- wizard, flying dragon, etc.
Still impressive of course, but not to the degree of what I saw on the marketing page.