Also: “think like real movie-makers do.” Scenes are shot independently, then “cut together” by the editor and cinematographer to produce the final, continuous show that the audience sees.
Frequently, the set for a particular group of shots will be configured with cameras in various positions (in a group of cameras), and lighting rigs (organized as groups), and various blend-files (and scenes within those files) which correspond to the various shots will “link to” those things as library assets. (In this way, the shot-files remain consistent, and changes to the library will be instantly reflected in them all.) The shots, all planned in advance, are shot in a non-chronological sequence, then finally assembled in a video editor and paired with a soundtrack.
The CG workflow has been described as: “Edit, then Shoot.”
The shots might initially be done using OpenGL Preview renders, which can be produced in near-real time, and these renders used in subsequent video editing before the final renders are produced. As with “real” movie making, more “takes” might be shot in this way than are used in the film, and they might contain a few extra seconds of action to facilitate editing. All of this extra footage, as with everything that “winds up on the cutting-room floor,” will never be known by the appreciative audience unless someone decides to put together a “goodie reel.” So, you do not try to initially render the “final cut” that will be the eventual product of the editing process … and you also do not final-render a frame in order to then decide whether-or-not you will use it. (If you’re not going to use it, you never produce it.) OpenGL renders by definition will perfectly match their “final” equivalents … and, for some passing shots, might even be “the finals.”