It's an interesting psychological issue that comes up in AI sometimes. There's a certain cachet to something being "online", in the sense that it's literally generating something on the spot, even if it's with a fairly simple process. But consider this transformation: take a generative art piece, pre-generate 100 billion output instances, throw them in a big database, and then the online code is: randomly select a piece of art to show the viewer. No actual practical difference, as long as you've pre-sampled so many instances that the user could never reasonably see any repeats. The space of variation is the same; in terms of what you get out it's just as generative as before, just with a different choice on the typical algorithmic time/space, online/offline tradeoffs. But the rhetoric is a much harder sell.