Discussion: what makes an AI creation a "gem"?

I run Slop and want to settle the judging criteria together. Here are my candidates.

1. You can actually use it: it runs, you can follow it, or you can copy it and use it right away.
2. A human clearly worked on it: the prompt was revised several times, the results were curated, or it was polished by hand.
3. There is something unexpected: at least one solution or phrase you would not have thought of.
4. The source is honest: it says which tool and prompt made it.

"Slop", on the other hand, is something you spend time on and get nothing back.

What are your criteria? Add yours in the comments and I will fold them into the site's judging guide.