It is still time you spend in prep. Now of course if you find it fun, that is not a problem. But if one doesn't railroad, you cannot always know what you need beforehand. Furthermore, if sorting trough assets is feasible, they are not limitless. Existing assets also are someone else's interpretation of how things look like and I like to personalise things and not just repeat the tropes.
Love prep work. But it really doesn't take long in Roll20. And everyone's imagination is finite.
I set up likely maps against future need; it takes 2-3 minutes to plug in a map. Ambush sites, city blocks (with every room laid out), a village, whatever.
You can personalize in Roll20. Add furniture, interactive traps, debris, bystanders, sounds (both interactive and background), changing terrain, battle damage...the list goes on and on.
And what I do is get scenario idea-find maps-tailor scenario to maps & accessories. thus, there's no interpretation at all: they see exactly what they are supposed to see. Much clearing than a GM droning on, and a player missing something while their thoughts wander.
Above all is dynamic lighting. That keeps the player discussing what they are seeing as each moves, and (because their PCs have armor mods that allow sharing images) posting screen shots.
That way the players are actually exploring, being engaged and interacting with each other and the developing scene, rather than listening to a GM droning on. And it lets the GM focus on the NPCs, rather than endlessly talking.
Using visual media such as VTT, the GM and players game
together, as opposed to a GM talking to an audience.