Everyone will draw objects/props/people/bits of the scene and I'll put them together. So at first it will be like the second option: there's mostly nothing to go off of.
However, I'll periodically post the in-progress scene image on my own submission page (I'll provide a few hours after the jam starts, when I start to put together sprites). So as the jam progresses, the scene will become more defined (and it will tend towards the first option you talked about). This is why I want people to submit their sprites as soon as possible :)
Thanks for asking about this. Maybe it's a point I should clarify, as it is soo important to the jam.