Combining facial rig / full-body mesh

[aaron_johnson]

i have a highres charcater rig (with bound skin) that needs a facial rig built into it (this wasn't planned initially but now i need it).

i can combine a facial rig easily enough by making blendshapes from full-body meshes in another file, but working with artisan on 40 or so full-body meshes is a bit of a hassle, i want to work on just heads in a seperate file. i know this is a common question and there's a way to do it by cutting off the head, making blends on the head, combining again but its not working for my rig.

here's the steps i've taken:

1. extract head faces from body skin
2. export duplicate of head to make blends in another file
3. combine original head to body (merge verts, soften edges, skin should still deform as before).
4. import completed head poses, creating new blend node on original skin (with check topology off).
5. re-order blend node beneath skin input node

ok, i can get up to number 4, and the head blending works ok, but it won't let me re-order the blend node in the input list... presumably because i've done something stupid or missed a step somewhere what happens is when the rig moves and the blend is applied at same time, the head will move back to zero.

any help appreciated, thanks guys.

[aaron_johnson]

no-one...? wow guess i'll have to go elsewhere to ask rigging questions.

i think i'll have to make head blends with full body meshes for this character.

[joojaa]

this is one of those stupid workflow things most people think is more optimal BUT realy Isnt.

Dont cut the heads its easier just to paint each one in the same fill on the full thing. I mena i often wonder about thsi becase theres no performance difference realy xcept the initial loadtime of the files with uncollapsed blends (yes most think its more efficient teh way you do it but for maya computatinaly it realy isnt)

THING is that you want to hide all the parts you dont want to use so to increse performancce so... USE ISOLATE Select on the head or jsut the face. Maya does not calculate points that arent seen.

Also theres NO benefit memmorywise for the blendshape to be lmitted to only those points if you use delete targets method as may aonly stores values that CHANGE.

By the way you cant order the blend to another part of the history because the blend is limitted to the subtree where the combine is!

But since you have allready done all the work do this:

Before binding blend each different shape on the full body and duplicate it for each blend then, DELETE HISTORY, recreate the blend with delete targets on the full body mesh rebind it and your now free to reorder as you please.

[aaron_johnson]

hi joojaa - yeah u confirmed what i was thinking, that breaking up the model and combining was causing the problem with relocating the blendNode.

i also agree on not detaching the head, more hassle than its worth i think, and a blend only calculates the verts that move anyway so this rig should work out good... i start work on the head blends today

cheers.

[sithwarlord]

another way you may want to try is this:

take a copy of the body delete all the faces that you don't want involved with the head blendshapes.
make all your targets and such with this new head, and create your blendshape deformers with these targets on a base version of the new head.
then use that new head as a wrap deformer on the existing full body geometry that is already bound. but after using the wrap deformer tool, you can then use the edit deform membership tool to isolate the verts that you need for the wrap deformer (the blendshape head) to affect - this will allow that the blendshape head will only be affecting the same corresponding verts found on the bound full body.
then you'll want to re-order your deformations so that the wrap deformer fires before the skinCluster.

my 2 cents at least.

[joojaa]

yeah well the workflow he had going was allready better then THTE workflow in terms of efficency.

[kubba]

ok guys, one dirty secret for single mesh characters..

my workflow would be-
1)duplicate body, delete faces you dont need for blendshape
2)rig the head the way you like it..
3)add rigged head as influence object to your smooth skinned single mesh character (turn on use geometry, polygon smooth 0 and lock influence weight 0, in option box while adding (you dont want to mess what youv'e allready painted, right?))
4)paint on head vertices to Head influence (all 1)
5) tun on Use components in your smoothSkin node

this is MUCH faster than wrap deformer + no waiting for wrap when opening scene..

[joojaa]

Well that wasnt realy the point. But yeah it works. But still your history has 2 extra items in it that dont need to be there. The influence object thing is same as a wrap deformer.

Thing is that all of those transference methods seem to kill the every idea of decapitating the subject anyway. Is tone of those workflows where you se ahey im saving memmory and doing less work. Whan in rality your wasting memory doing it slower and wasting time on a thing.

Theres nothing to gain by decapitating the mesh. It may seem like a smatrt idea. But it isnt. theres no gain in it. No savinhg nothing. whats so hard to get about it. maya treats the full body sculpt way as lighter in the end over all those world influence transferance methods. So why even bother going the hard way around end up dooing a rig tahts heavier, or that needs baking when the real deal was there in the seemingly (but ist only a illusion its no heavier) heavy workkflow anyway.

[kubba]

no offence man, but:

Cipher wrote>
"40 or so full-body meshes is a bit of a hassle, i want to work on just heads in a seperate file"

so, well that really was the point. Yeah it works. No extra items. Not really, its faster.

guess Cipher already sculpted some of his heads so 1 minute more is way faster than redoing blendshapes on full body..

personally I like to manage head and body separately, and some more load is not a big deal, especially when animators can avoid evaluation while animating..
( common pipeline I usually set for animators is animate with proxy, than when lipsync needed turn on just influence head so theres no connection to highres geo evaluated)

in your case, animator would have to turn on highres geo with all the deformer stuff above to see facial animation. And man that would not make you lots of friends...

[joojaa]

Im totaly aware of how to build animation proxies. The thing is that theese are different strategies i realize it, and all fine but the thing is IM JUST POINTING OUT that its not as ovoisly bad as you and many others seem to think it is.

Also this illustrates why cartain toolsets appear to be missing for managing such things more elegantly. Tough thats not allways realy true.

so, well that really was the point. Yeah it works. No extra items. Not really, its faster.

Well you could think that, but before you say so consider this:

It raly isnt, thats my point. You think it would be but Maya has this nice feature called ISOLATE SELECT. Now if you take a full res body then isolate the faces you dont want to see ie only the head. You doing exactly the same thing only the original data is intact, and there but maya does not calculate it or process it. (now to get the full power of this you may need to isolate on all views but still...)

Its only minusculely slower than the separated heads. BUT carries a lot less hasslle around. prettymuch the only thing significantly slower is file load and some ops that work the entire body... That said it was a final modell so there wont be any of those around.

SECONLY if you must separate the heads for modelling because thats realy realy the way you must bash your head trough the thing. You will want to avoid the wrap and influence deformers, they are SLOW especialy on highpoly things the cut and merge techique is far superior in many ways. wraps are not not only slow once its slow all the time even tough they have made sigificant speed iprovements on them.

But thet said theres a trick to all this wich redeems both wrap and influence. Once youve made your head fully you can take it and wrap or influence the full body and make full body copies of each blendsahape snipet and then collapse the deal to a single singele mesh with one blendshape input.

Now this is significantly faster to animate and use then the full body wrap/influence version (okay if you only ever use proxies its a minsucule deal but as it

[kubba]

I'll make it quick as I'm getting bussy now..

1) combine 2 skinned objects, merge vertices, smooth,
try to add blendshape.. no way ? ok.. delete smooth, merge vertices and combine node... its fckd? yeah...

I'v already tryed this way in real world production and due various reasons it didn't worked well (the one above is just one of them..)

2)nice thing about influence is that it can be skined or whatever, so it moves with skeleton without double tr, what's very different from wrap (unless you use relationship editor for 0/1 relation, of course ..)

3)well you've just revealed me a crazy new stuff -preventing display of several faces in viewport says maya not to evaluate all the DAG stuff... ?

please before you write anything, try it ..

[joojaa]

1) i dintnt sqay it was optimal... but it works! it actualy does. And in a real procdction setting. BUt i didnt recomend it is just said its better to bake out the entire thing to full body mash blends. Besides whats wrong with one smooth node and one merge verst node (yes many same node ops can be collapsed into one after the fact)

THere is one problem with the bound head and thats that the vert order becomes different, but taht too is realy easily handled. IF your not limitted to maya default tools.

My entire point was NOT to ever separate the ting at all. since theres nothing realy huge to gain form it in most cases, except more work. It is offcourse a bit slower tahn the separate head since it does imue a bit of extra overhead but still much faster tahn the whole set!

offcourse i might be a bit biased to say so as im realy running on tip nitch hardware, so i rarely need to sacrafifce workflow for speed.

2) yes i can se that influences can be hanider than warps. Wraps offcourse do have soem other advantages. but basicaly tehir the same piece of code. As is clusters and smooth skin.

3) maya will not eveluate things it will not need to show if they dont influence the the views. So if you hide part off the object then if the deformer does not necceserily need to update all the points. This is the most basic core bahaivour of maya, the essence so to speak.

This applies to DG as well as DAG (DG offcourse is what defines history of thinsg usualy). Things are not push torough maya but rather push pulled. wich means maya does send a notification fo dirty upward but the upward node decides whan it needs the data and what part of the data it needs. Most maya internal nodes indeed do only process whats needed os if the points are hidden they dont evauate, untill asked for.

Offcourse if a hidden node affects something esle it WILL update but only as much as updating that node downstram needs.

Now this can be easily demonstrated witha a polyproxy node and the subd shape. Even is the subd is slow bacuse its huge the polyproxy will still be much faster if you go and hide the subd sshape even tough its not

So to mekae things totaly clear i was

A) refeering to his situation. whare hed allready sewed the decapitated head back.

suggesting him to bake.

C) suggesting him not to have done so in the foirts place.

[kubba]

im slowly loosing point of discussion, so shortly:

no matter what you say, isolate selected will not make it run any faster

please try it:
make a spere say 100 100, smooth bind it, animate skeleton, write down fps..

now select one face, isolate selected, write down fps...

compare values... any difference?

unless you hide whole object it will always evaluate all the history stuff.
I'd be very surprised if Alias rewrited all deformers when they added Isolate selected function as it was not there since very begining...

[joojaa]

62 fps and 62 fps. Thing is i need a asolumutely huge modell to show any signifivant slowdown in performance! BUt i am running a quaddro 4000.

But no it does not evaluate all history at lkeast not to me. Ive even run degtimer to chek this. Altough selection does not seem to speed up on isolate!

[joojaa]

OH godamn... Appers that my isolate select function is a bit rejigged. Theres a host of code behind it that actualy does the same thing as separate head except on fly.

Whan i disabled it you did actualy not se all that dramatic performance boost. Altough for your example thers still no performance hit.

[joojaa]

Nice one... It actualy deletes the verts with a delete history node pipes it to shape for caching and whan you want the original back just delete theese 2 nodes.

The intersting one is acytualy how it handles renumbering! It uses a mysteriou reorderDataNodes. Ill digg for its API code now to se if i can find it (would be nice to share but im not sure if i can).

if somone wants to looksee what it actualy does ina faked way (it works biecuse ive not actualy needed to reorder in this special case, thats what the mystery node is ther for).

{// to faciliate deletion autoamaticaly
itemFilter -bn "Fix*" Fix;
delete lsThroughFilter Fix;
}

heres a file to dowload Note the faster sphere turns complete again when you delete the fix nodes. Also note its faked becuse without the reorder it wont work on arbitrary meshes. but still the solution is nifty to say the least. Never noticed it was there

Alias could have done this by the way its actualy usefull. Note soem users will forwn on the solution because it tampers with history but it meakes my life so damn much simpler

[kubba]

ok enough:
if you'v runned dgtimer you could see that there's no difference between isolated and non isolated results => no gain (in fact isolated is tiny bit slower)

haha no matter what GFX card you use deformer evaluation is CPU based not GPU, unless its vertex shader, which is not maya's deformers case (you could see maya running on TNT those old days, right?)

so not even "I'v got oxo 9000 card" sentence will make you right

[joojaa]

I wa sworng i admit it, but read the nodal idea/solution its actualy quite interesting (i never actualy noticed it was there garntted so i started to thing the whole world was mad). read the 2 posts i crammed between whan you were typing that one.

I must say that analying this reorder node si a bit hard to say the least. Its coded but not commentted UCK.