Automated inbetweening

[sykosan]

...

[idragosani]

I hope this will be convincing enough.
I would appreciate not to hear again that this is too CG or too smooth, because such comments would be missing the point.

It's not that it is too smooth, but it is very mechanical. What if, instead of just this literal tweening between first key and end key, I wanted to have the head dip down slightly, with slight ease in and ease out? Would it require more keyframes for your method? This almost looks like one of those graphical morphing apps....

[Fabrice]

In terms of rendering, it looks like it's very close to a mix between classical morphing of lines + more or less a wrapping grid + some fadings.
In this situation the preparation of each keyframe can't be less than 20 minutes, which is a lot, then you still have to perform a clean on the top of all inbetween.

I'm curious, did you already play with the FX-Stack of TVPaint ?

[ZigOtto]

yeap, it appears as a combitation of Warping FX + Dissolve,
for each seperated element, you warp keyA to fit to B place, and KeyB to A place,
then you cross-dissolve the two warped sequences.
that's why in this example, the nose becomes semi-transparent in the middle of the move.

[Paul Fierlinger]

This would have to take S.Tiger 30 minutes to draw 12 inbetweens resulting in a far more pleasing performance. What we see here looks like an animated motel sign silently glowing in the dark.

[sykosan]

...

[Elodie]

Why can't I see the result ?
I was so curious...

Thanks everyone for the comments.
Time to move on.

Okay... you post the result at 10:24 PM and you withdraw everything at 8:00am the day after ?....You got 3 opinions and a question from Fabrice (how lucky he was to go to the forum so late !) and then, you disappear and withdraw all the links to your job ?

I'm perplex...

[Klaus Hoefs]

I'm perplex...

To be honest, I am not.

Didn't it sound that impatient way from the beginning ?

----
Anyway, there is nothing to say against of experimenting with automatization and to have a look at this from the more technical side. But better to face the results and to fight out in the open.

[slowtiger]

Seems I missed it. Did anybody save the example and can post it again?

From Fabrice's and Zigotto's description it sounds like something I indeed had seen elsewhere last year, it was announced as "the future of animation" (as it always is) and showed some anime-style head being turned around withing a 30°-limit (from 15° right to 15° left) - exactly the kind of view that would be very easy to do in such a fashion where parts are arranged by Z-depth and then moved at different speed (parallax). I chose the 90°-left view on purpose (ain't I mean?) because the biggest unsolved problem with all automatic solutions is how to deal with lines or parts which are only visible in one of the extremes.

[Elodie]

Sykosan contacted me and does not want to answer this topic anymore.

[idragosani]

Oh well it was fun while it lasted. Head turns are notoriously difficult to do with 2D vector animation. If you go on the Anime Studio forum and ask how to do one, you will be told to use layer switching and essentially have to do frame-by-frame animation to accomplish a good head turn.

[Elodie]

Head turns are notoriously difficult to do with 2D vector animation

Humm, not that much IMO


But yes, the time you spend to put the grid and move the points are really long, maybe longer than inbetween handly made (or Sykosan's technique, maybe ? We will neve know)

[slowtiger]

I really don't get it why some people are so obsessed about using a hammer as a saw - or 2D animation software for mimicking 3D stuff. Part of the fun of course is the "because I can" aspect, and I bet each of us has done stuff for that reason at one point in time. (I did, maybe I'll post that stuff one day.)

But if you absolutely need slow turns of complex objects, use 3D! Less time wasted. You still could draw over the result to maintain your style, if the shaders don't do that. Disney animators used maquettes of their characters for reference, or even photographed 3D objects (like carriages and stuff) and traced the prints. Still nobody accuses them of cheating.

And another point: there's a reason why usually we do full head turns only very fast - or, if we do very slow ones, then only within a small angle: because this looks best in our medium, "drawn animation". Demanding slow-motion 180° head turns is the equivalent of asking for a 30 second leap in classical ballet: it's not within the toolbox of this art, it could be done with rope and pulleys, but frankly, who would want it?

[idragosani]

Head turns are notoriously difficult to do with 2D vector animation

Humm, not that much IMO


But yes, the time you spend to put the grid and move the points are really long, maybe longer than inbetween handly made (or Sykosan's technique, maybe ? We will neve know)

Yes, but the angle of that turn is probably less than 30 degrees (and probably uses a 3D model?) If you want to do a full head turn from right profile to left profile, or even right 3/4 to left 3/4, and make it look fluid with ease in and ease out, possibly a slight dip (rather than a mechanical on axis turn), you are getting into some difficulties because of hidden lines you have to reveal. It *can* be done with pure morphing, layer re-ordering, blah blah blah, but with all that work involved, it'd just be easier to work out a frame-by-frame sequence... or rotoscope a 3D object.

[slowtiger]

Ah, I finally found the video I had in mind the whole time:

http://www.youtube.com/watch?v=YrLnF7CQ8Ac" onclick="window.open(this.href);return false; (there's a link to the makers in the description)



Forget that real-time part - there's no way to convert any 2D drawing into what's needed for this in real-time. Truly the conversion was done beforehand - you can clearly spot the different levels of the broken-up drawing.