Next-gen Facial Animation using Bones

Snowfly

http://www.satoworks.com/masterclasspage.html

Too bad the artist doesn't show his setup.
Still the most memory cost effective way to animate a head? I would think blend shapes are quicker to make and setup, but are more expensive to store in memory. I'm guessing current and next-gen systems have enough muscle to store blenshapes though (some characters in RE4 had 30+), and they are promarily used in cutscenes anyways. I'm not an industry pro..Anyone care to play devil's advocate?

monster

I don't really see anything that qualifies this as "next-gen" other than simply adding a few more bones than usual.

Blend shapes use more memory, and are usually created from a bone rig anyway. Bone rigs use more cpu/gpu. This used to be a bottle neck, but now that modern gpu's can do hardware skinning there's almost no expense.

On all the games I've worked on, I've had an easier time with the ones that used bones instead of morph targets. Programmers usually want bones in the eyes anyway to target objects.

Daz

I know Sato and have worked with him. Nice guy, very talented artist but a baaaad Art Director. Sheen I know too who did the rig. Very clever guy. They're both at EA LA. The rig looks to be based on the Jason Osipa method?

It's cool, but like monster says, there aint really anything special going on here. I hate the stretching going on with the upper lids UV's in the last shot. Personally I think you get much nicer results from morph targets. You have more control over creasing for instance. Try doing a convincing smile using bones. Now try one with a morph target. Much easier. But If you can seperate the creasing out into animated bump or normal maps, cool. The obvious caveat with morph targets is all that time required to sculpt those shapes. Plus of course storing them in memory. So no, I don't see how blendshapes are quicker at all. They're extremely time consuming. I've worked on a game where we had 35+ shapes to make per char and it was a royal pain in the arse. Geo changes required to base head after the fact? Jeez, nightmare. Bones are good, but you need quality weighting. Without that it all falls apart at the seams.

This is good quality stuff, but I don't see whats so 'masterclass' about it? Fairly high res head? Lot's o' bones?

Snowfly

The Osipa method -- could be. I've found that a facial setup with bones lends itself naturally to muscle shapes/visimes, rather than specific phonemes, and creating the phonemes is just a matter of saving out combinations of poses to Set Driven Key or whatever.

I am a little frustrated with weighting the face.. Component Editor does make it sort of fun though. So yeah that's where the attraction to blend shapes begins... Never thought about it in a production environment though. 35+ shapes per character.. ugh. I can see how a joint setup would be transferable to multiple characters with the same topology in the head too.

So inside Maya you're free to move, rotate, and NU-scale joints, but would you be able to get away with this in a game? Or are they rotation-only?

Faucet

http://www.jasonosipa.com/Downloads/Movies/NVidia%20Demo%20Movies/

Here's another nifty realtime face animation thing from Mr. Osipa. Basicly just texture blending. I imagine with normal maps and stuff it'd look really nice.

monster

Faucet, now that's a better "next gen" example. I've been using that method for years, for wrinkling clothing, but never in real time.

Snowfly, most games that use bone animation support non-uniform scale, move, and rotate.

Eric Chadwick

If your system supports it, you could combine blend shapes and bones together, for the best of both.

I'm finding ittough to get the bone pivot positions right, and weighting is time-consuming. Anybody have any tips about these? I know good topology is essential. But with the weighting and boning is it just a matter of experimenting and testing and refining...? Seems like it is.

Snowfly

Seems like it would be nice to combine both bone and blendshape animation on each vertex. That could reduce setup time I think, if you could roughly place and weight to the joints, and create 'corrective' blend shapes for the poses. dunno, maybe it's just better to get the weighting right, but some areas may be easier to set up with it. (i.e. cheeks bulging on a squint, collapsing when the mouth is stretched open...how would you set up a pucker??)

The texture blending example, would that require that you save out the crease for each 'crease area' into a separate image (like Osipa does it in the book)?

diZzyWalnut

[ QUOTE ]
http://www.satoworks.com/masterclasspage.html


[/ QUOTE ]

That's very impresive. I will use this topic to ask You if You know some nice tutorial about "facial animation".
(under 3DS MAX or MAYA ).... I would like to learn it one day.

Thank You

Faucet

The topic of realistic facial animation could easily fill a book... so erm... I highly recommend this book.

http://www.jasonosipa.com/JasonOsipa_Main.htm

It goes over everything from propper topology (extremely important) to setting up those complex rigs.

CheeseOnToast

Yep, I second that recommendation. A great book written well, even with a bit of *gasp* humour.