|
Bernard Lebel wrote:
Hi,
We're about to start layout on a 45 minutes full-cg show.
Sweet.
A few of us here are debating the idea of doing the majority of the
layout-previz and animation in a "master" scene. Such master scene
would include basically all the cameras and animation for an entire
sequence.
There are several motives behind this approach:
From the tone of your email, it sounds like these motives are coming
from someone else besides you, and you are looking for good arguments
against doing it. If I'm correct, I side with you on this one. Master
scenes of that scope are not a good idea.
- To enforce continuity between cuts. The idea is to be able to
playback the scene and have all the action take place more or less in
one go. So we make sure, for example, that the character's hand that
was at a given position on the cut out is not breaking the continuity
at the next cut in.
Here's a proof by contradiction of why this isn't a good way to work.
If you had a master scene that you could essentially play back as a
continuous animation, you could very quickly animate the world's longest
continuous camera move. But the world's longest continuous camera move
can't be easy to create -- it takes a while just to get a decent 30
second move. So, there must be something really difficult about working
this way. QED.
I think that the biggest problem is parallelism. Even if you could
easily shift actions to the left or right of a sequence to account for
allowing more time into the sequence (which I think would be a huge pain
in the ass by itself), how do you coordinate all of your artists working
at the same time and making such changes. I guess you could create a
concept of sequence times and come up with some sort of data structure
that allows people to work on a single sequence and have scripts that
automatically build the master scene, placing animation in the right
place, based on the sum of all the sequence durations that come before
it. Basically, you end up needing an editing program to move your
animation around, so that you can break things up into chunks and still
control how it all fits together. In itself, this would be really cool,
but it's not something that just works nicely in XSI already. If I were
to suggest an approach, I would say you'd want to do all of your work in
animation clips, and store metadata about where they go relative to
sequence beginnings, then have scripts that can go through and collect
all the clips and apply them in the appropriate places or something.
Lots of R and D would be involved, I think. The end result would be
awesome and very powerful, but it's certainly not going to save you time
on the first show.
- I have good reasons to think we won't have good old-style
storyboards, or complete ones. It seems the plan is to have the layout
artists do full-fledged storyboards in 3D, in order to reduce
continuity problems but also to make sure things are always at the
right scale in the FOV.
Now, I can see an interest in doing this for the layout/previz stage,
but I have a hard time figuring how advantageous such an approch to
animation can really be.
I'm not the best previz artist in this place, but I've watched enough
projects to know that storyboards are always a good idea. When we don't
have them, we make them. Sometimes we even make them in 3D, but we
don't forego boards in favor of going straight to previz. Instead, you
want to block out a sequence of shots that makes sense first and worry
about the animation of it later. Unless a shot only makes sense when
seen in 3D, rough pass 3D animation just gets in the way. Plus, people
tend to get married to it and the sequence suffers. You just want the
quickest, dirtiest concept of what the shot is, so that you will feel
free to keep changing it until it's good. Then, once you're happy with
it, and you think it has a good shot at making sense, you start doing
more animation.
For previs, we do, as often as possible, keep a single action in one
scene file and cover it from different cameras. That's sort of a
no-brainer when things exist as pure animation and you're not worried
about any of the weird end-cases.
Continuity is a strong argument in favor of this workflow, but.....
- Accounting for "cheats" introduces some hazards. For instance, it's
not uncommon, to fit a specific shot, to move props or even characters
from their "standard" location to another location, simply because the
director likes it better that way. So it means that the animator might
have to animate the position of props, stage elements and even full
character rigs. As a result, some objects may just jump from one place
to another.
Typically, this is done using reference models, right? So, your scene
can contain the information about where things need to go that's
different from the default. But you can still choose to leave the model
untouched, so that things naturally fall into a position that's
consistent with the rest of the sequence. Again, I could see
engineering some system for doing this within a scene, but why not
leverage the tools that are there already?
- It is my understanding that shots that have requirements too complex
to fit in the master scene would become separate scenes. So again I
foresee a nightmare of trying to keep track of what part of what
master scene has been done separately. Worse, what if we want to make
modifications to the master scene, and then pipe these changes down to
the specliazed scenes? I mean it certainly is possible, but then it
need to be extremely tightly tracked, right? Lastly, from my
experience, what everyone thought to be an exception or a special case
actually happened to be quite a common scenario. I'm probably
paranoid, but getting bit by an inadequate workflow is the last thing
I want.
Again, reference models provide a similar function already. But it only
really works in a multi-scene workflow, where you're free to mess with
the reference model differently in one scene from the next. Maybe you
could do it in a master scene, but it would totally defeat the purpose,
right? Of course, you still have to keep track of one-off shots that
need to break referencing. But I don't think there's really a way
around keeping track of that. In the past, for simple
reference-breaking changes (such as deletions), we've maintained scenes
with a reference model and an embedded script to repeat the change from
the original ref model and re-export the modified model. We call them
autobuilds. If you maintain a tree structure that links your one-off
models to their originals, and knows where to find the autobuild, you
could write a model export wrapper that updates the one-offs too.
- That would also mean having animators (or else) "slicing" the
animation and export it for the "special" version, and possibly
migrating the "special" version back in the master scene. This is
entirely possible by using the Mixer as a "montage table". But again,
I see big red signs that say "PROBLEMS AHEAD", mostly because of human
errors.
Yes, this is much easier if the animation lives in smaller chunks that
don't have to depend on placement in a global timeline. Things can
still have a notion of global position if you want, but they should all
reference an offset for the sequence or some smaller chunk that is
manageable.
- Our characters have syflex simulations. Obviously if characters are
jumping, bad things will happen. So we'll need to have some "rollout"
frames between cuts for the simulation to get back in a stable state
(unless the character is moving violently, of course). I guess this
problem also applies to motion blur and various time-movement
dependent effects.
The way I see it, this falls under the umbrella of a concept I've been
thinking about lately of shot-switching. What if XSI had a shot
selector, just like it has a pass selector? This would be useful, no?
It would make it much easier to maintain a scene as a series of actions
that get covered in a variety of different ways, from different cameras,
etc. Maybe the best way to do this for now is to maintain a database
outside the scene and run scripts that change things around?
- Then if we have to move objects around in the same shot, and we have
to put delays between "action" time, then I see the whole thing
becoming a nightmare in terms of management. It means someone will
have to keep a very, very tight look at frame ranges and make sure
this info is propagated.
Yeah, the only way to do this without it becoming a crazy nightmare is
to use some sort of implicit timings. Otherwise, they timings for the
guy working on the last shot of the film are affected by what everyone
else in the office is doing. It's much simpler if he knows that all he
cares about is where the stuff fits relative to the beginning of an
action and/or a sequence.
Given all the previous points, where is the advantage of doing
"master" scenes? I'd be interested in people who have tried this, if
they have some thoughts to share.
I think there are things you could potentially get from doing a master
scene that are good, but actually doing a master .scn file is bad --
especially because it doesn't work nicely with the key non-linearity
tool you have at your disposal, referencing. Having the concept of a
world, and time would be good, such that if you wanted to, you could
hypothetically automate the creation of a master scene. Depending on
how big your world is, though, a master scene might not even work at all
if you can't recenter the origin. In that case, it seems like it might
be a good idea to maintain an offset for the scenes, so that the central
action of the scene can be placed near the origin, even if it would have
been far away from it in a master scene. This will help avoid some of
the MR artifacts that we've seen in assets as small as 2 miles (with XSI
in feet). This is of course assuming you want your models to have a
sense of where they belong in the hypothetical world -- otherwise,
there's no point.
Thanks
Bernard
---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in
body:
unsubscribe xsi
---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
unsubscribe xsi
|
