Using live action as a reference: (Reference: CGChar)
" I think live-action reference can be incredibly valuable as a tool for analyzing both physical and emotional aspects of a performance. We use it like crazy over at ILM - especially because our characters are going to end up interacting with live-action characters, so we try hard to get them to hold up under those circumstances (sometimes more successfully than others!)
Anyway - as far as how literally you should "copy" your reference.... Well, part of that will depend on the style you are going for. I know guys over at Pixar who are using reference all the time in their work, so it isn't like you can say "well, I don't want to use reference because I do cartoony animation." That's crap. It's always worth recording yourself or a friend performing a scene to at least see what would have happened in real life, and taking from that whatever you can.
Even "cartoony" work needs to be grounded in reality (both in physical motion and in a believable acting performance) for an audience to identify with it. I think that's really important to keep in mind.
Sorry - got a little sidetracked. Okay - so you filmed reference, now how much do you "trace" it. The previous poster is correct that the Disney guys found the rotoscope stuff to be lacking a lot of the vitality, energy, and "life" of their regular work. That's actually something that many companies have also discovered about straight mocap. Same principle, different medium.
I would recommend filming yourself performing the scene/dialogue TONS of times. Perform it until you no longer have to think about the actual words the character is saying, or the rhythm that the character is saying the line in (if you're using a pre-recorded line). Record yourself until you reach the point where you don't have to think about that any more. THEN you can start to get into that character's head and think about his "internal monologue" and the subtext of what he is saying more than the actual words. You need to become the actor in a scene like that and truly feel the emotions as honestly as you possibly can in the scene.
Once you reach that level, film a few more takes that are honest takes. If your scene is a line of someone pleading for his life, get to the point where you aren't think about the words he says and are thinking more about the emotions that guy is feeling - the panic and sadness and frustration and fear - and then just see what your body naturally does.
Okay - so then you've found a solid performance that is grounded in reality and that people can identify with. For me, the next step is to find my key poses and breakdowns in that performance, and for me, there's a LOT of them. I rarely would go more than 5 or 6 frames without a key or a breakdown in my blocking (remember, the computer is the world's dumbest inbetweener - you need to hold its hand). And a lot of times, for my blocking, I'll just take the literal frame count from the reference. Pick my favorite moments, my key poses and my breakdowns, draw thumbnails of them and mark down the frames they really happened on. (if you're doing 24fps, don't forget to convert the frame-numbers from the 30fps reference).
Now you get to do the really fun stuff! Next, I take those thumbnails and start layering the principles of animation on top of them. I'll push the poses, try to make them more dynamic. Exaggerate this arc, push that line of action, exaggerate a reversal here, dip the head a bit extra there, get a nicer sillhouette, etc.
I'm a terrible drawer. I'm crap at drawing. I used to be only medium-crap at drawing, but after nearly 6 years at ILM in front of a monitor only having to draw thumbnails, I am now plunging into the dark depths of "full-crap." But that doesn't matter. As long as I can draw a thumbnail of a guy that describes his hip position/angle, the angle of his shoulders, and the position/angle of his head -- you're set! A stick-guy like that can be used to thumbnail damn near any character you can think of.
Do yourself a huge favor and figure all this stuff out on paper FIRST before you ever save a key. Ideally, by the time you touch a mouse, you already know exactly what pose happens on what frame. You already figured out your paths of action and your arcs and anticipations on paper - then just get into the computer and plug that stuff in!
Once I get those poses and breakdowns into the computer, my poses might be exaggerated, but my timing is probably the same as my reference at this point. For me, now I'll go in and layer the animation principles onto my "timing" in the scene. Punch up the rhythm of the actions, make sure things aren't happening on these metronome boring evenly-timed moments, find places to push an anticipation or add a bit of extra overlap, speed up an action for effect, etc. For me, this phase is easier to do in the computer than it is to do on paper. When I block it in, I'll save a position on every possible part of the body for ever key and breakdown pose - then to move the timing, all I have to do is grab all the keys on all the curves at one frame and shove 'em over.
Once I have the timing down, I'll get in there and start the polishing phase - checking my arcs, making sure things read, offsetting different body parts, checking overlap, etc.
If anyone ever tells you that live-action reference is cheating, they're either not a professional animator, don't have much features experience, or they're feeding you a line of bullcrap. It's really important, and you can bet that you're favorite animators, whether they work at Weta, Pixar, or Disney, are all using it, at least to some extent.
That's probably a really long answer to a simple question, huh!?
Short answer: don't copy it, but learn everything you can from it. Layer your knowledge of animation principles on top of it. And do as much as you can in the planning stage first. My best scenes, and the ones I finished the fastest, are the ones I spent the most time planning... "