Ya, The maps are 2D images, where pixels are geometry and the alpha map is open space. When I import the image, I do so as a large sprite, then apply a polygon-collider. So in that sense, yes, the game interprets the picture as one big level.
However, it is not very precise, and I end up with about 300-400 vertices for the collision mesh, where I need more like 500-700 for smooth curves etc, that is where the majority of the work comes in, adjusting the mesh to make it playable.
Luckily I can just move the mesh vertices around in real time so it's just a bit of a grind to get the map looking nice and feeling right.
But if there are other things like moving platforms or bouncy floors, I have to add those in as well, since there is no way to tell a picture to add a moving platform here or change friction for a certain color (yet ). So the difficulty of the level process just depends on how big and complex the level is.