Specifically, what I suggest is this: start by “blocking out” your world … and by “blocking out” I literally mean, blocks. Simple shapes, the roughest modeling you can manage, but “to scale.” In other words, when you superimpose a long-shot from your finished film with the blocks shot, in an ideal world it would almost match.
Now, get down there into that world and start imagining your movie. Insofar as long shots are concerned, put the major shots “on the map.” Then, start with also-blocks rough sets for the closer shots, also to-scale. Always include a human-sized shape in the shot. (Sphere, cylinder, two cylinders for legs, two for arms. Done.)
What you’re doing is working out what each shot will be, and that determines where you do and do not need to be concerned with modeling of any sort. If you do not determine that you need it, it doesn’t get built. Lavish attention based on how much is or isn’t seen. If you can’t see it, don’t decorate it. If the detail won’t affect a substantial number of pixels, don’t add the detail.
When you get to filming individual scenes, you don’t know in advance what’s going to be the final makeup of cuts. So set up different fixed camera-positions, including dolly tracks and booms and so-forth, and shoot animatic-style “film.” Then edit that. This guides your refinements. This tells you what frames you need. This locks in what you can and can’t see. You make very fine decisions about f-stop, camera angle, moves, lighting and so-forth using OpenGL renders and to-scale geometric shapes. You are going to lock in those decisions; you are going to make them now. You hold up that geometric mock-up against the final and it matches.
You know that the time-waster on any and every CG effort is render time. So you absolutely do everything that you can, first, to minimize that; then, to do it in such a way that whatever you do render can somehow be kept and used.
Real-life films are (sort of…) “shoot, then edit.” CG is “edit, then shoot.”