If you adopt the title of this post as a mantra, it will make you a much better CG filmmaker. Your movies will look better and you will spend less time (and more-productive time) rendering them. The core idea is a simple one: “first, shoot lots of film, quickly,” then “leave most of it (who cares?) on the virtual cutting-room floor.”
(In this piece, I am only going to describe the workflow at a very high level. This is not a step-by-step “how to.”)
We’ve all seen amateur CG films which have “one long take.” The camera doesn’t move at all while the hero walks up to the door, tries the knob, finds it locked, peeks through the window and finally walks away. But you know from experience that a real movie-director would never film it that way: this action would instead be depicted as a tightly-cut series of shots, with a mixture of framing and camera angles.
When CG artists realize this, the next thing they often try is to still do “one long render,” but to animate the cuts during the rendering process. Not only does this take just as much time as before, but you can’t change it after it’s done.
A far superior process takes advantage of the OpenGL Preview feature of Blender, which uses your computer’s hardware to quickly generate frames, and the Stamp feature, which can stamp each frame with things like the file-name, camera and scene name, and frame number. Now you have the ability to “shoot footage quickly” – a minute’s worth of footage takes about a minute to produce – and the footage will exactly match a “final render” of the same thing.
A key to this technique is to make aggressive use of Blender’s library linking feature, both between files and within them. Each set, actor, and prop is a linked asset – as a named group. Each shot is a “linked scene” (prior to Blender 2.8; in that release, the data-structures are different but the ideas are still there). You need to be able to reproduce any shot at will. You need to be able to edit an asset only once, and see those changes in every shot where that asset appears.
Key Point: Although you are “quickly shooting footage” now, each shot-file – if it makes the cut – might become(!) the finished one! “Stand-in” props and other assets – if they survive – become the real ones. Sets, now properly-dressed, become the real ones. (That’s why these details matter.)
Another important consideration to establish at the start is physical scale. How many feet/meters tall is that actor? How long is that gun? How big is that set? How high and wide is that window or door-frame? Everything must be correct and consistent. Get out that tape-measure!
Finally, you want to defer every detail that you can, until you have to do it. For instance, actors and props can be made of simple geometric shapes (placed into individual Blender groups for linking purposes. These must be “to scale.” But why detail-build a gun or a wheelbarrow when you don’t yet know if, or how, it will appear in the show?
Once you’ve set things up as described, start creating scenes. Place several named cameras in the base scene from which to shoot the action. (Use “Scenes” to separate the outputs.) Film the actor stepping up to his marks, doing his action, then walking off, to give you “tails” for editing purposes. Now, shoot film with reckless abandon! If you “might” use it, go ahead and shoot it. (Unlike Waterworld, no producer is going to come down upon your ass for shooting too much plastic, and your virtual actors don’t get paid “union scale.”)
Remember: "if you want to change it, make another one, instead."
"Actual film-making" occurs now, when you take this footage to a video editor of your choice. (Blender includes one, but there are others, notably ShotCut and OpenShot, or iMovie, which comes with every Macintosh.) This is where you begin to assemble the footage that you’ve shot into your actual movie … and, once you begin to get the hang of it, you’ll discover that it takes a lot more time than you suppose! However, you can actually move your project all(!) the way to final(!!) cut in this way, with actors that are still “cone-heads” shooting with cardboard-box blasters at cylindrical bad-guys.
You can work out a scene in a game in exactly the same way.
You’ll find yourself “rolling” between two shots looking for just the right cut-point, considering the rhythm and the pace of the sequence, and making all kinds of decisions now that might never have occurred to you. However, because you can generate (and re-generate) the footage quickly, you can afford to do this, and to do it well. This is a key creative activity – the most-important one of all.
Plus – I gotta say this – "this is fun."
You’re not “on tenterhooks,” waiting for hours to see what your movie’s going to look like, wondering if you’ll just have to do it all over again. No, you’re assembling your masterpiece now, and if it suddenly occurs to you that a new shot is needed to (maybe) fill this-or-that gap, you can switch-gears and go make it. And if you’re not yet sure “which path that diverges in a snowy wood” will turn out to be the right one to take, you don’t have to decide yet. This is a liberating feeling: “you’re still the director, you’re still in charge of this thing.” (Oh, damn … what time is it?!)
As you get closer to final-cut, you can begin to see which props actually appear in the film and which didn’t make the cut. You can also see how important the asset is; how much time-and-attention needs to be devoted to it. (Remember: the “towns” in spaghetti Westerns were only skin-deep, but the horses were real.) As you replace the stand-in assets with the real things, keeping the Group-names the same, they magically appear in every shot due to the power of “linking.” As you place the real lights on the sets, they appear both in the 3D Window and in your previews.
The final step is almost an anticlimax: do final renders of the exact frames that you already know you need, and drop these one-by-one into the final-cut edit. Right before your eyes, your “mountains of footage” become your final picture … or, pictures? … one shot at a time.
The result of this process will be a much more realistic movie, cinematically much stronger, and you will feel very confident about the upcoming result long before you hit the “Render” button for the first time.
2020 update by the author:
Some of the terminology used in this piece has changed yet again, and the “OpenGL Preview” rendering option has been replaced. But it is also much less important: the EEVEE renderer can run in near real-time while producing excellent results, and also now drives the modeling window. Also, the “Workbench” renderer operates in much the same way that “OpenGL Preview” used to. (But is also suitable for “real work,” and might be your choice depending on the project.) The workflow discussion remains the same.