Face Topology [Breakdown Guide]
polycounter lvl 9
~ (NOTE: I wrote this to help some students I know and some PC friends mentioned posting it on here to help out the early learning community. Feel FREE to dive in and we can start a fresh discussion on the age old face-topology subject 
)
)Students seem to often find face topology so alien and hard to learn so I decided to just label out the basics of what I found to be useful rules when learning to get functioning topology that can deform well, and helps sell the forms correctly. So if you find this helpful thats great, because I sure wish somebody taught me this shit when I was learning instead of just sending me a link to a super dense wire-frame of a well known game character and saying "Do it like this. This is good" etc. On that note; less talking more images.
The method I use to make heads is edge modelling instead of starting with a box. Which is essentially where you start with a plane and extrude out from it sequentially, but whatever method you use is up to you. This post is just to talk about why I put my loops where I do etc. If none of this makes any sense whatsoever that's fine. Just be a piano instructor or a stuntman instead.
[ame="
http://www.youtube.com/watch?v=nxcOTVuXcUI"]Click Here For An Example Of WTF Edge Modelling Looks Like[/ame]
http://www.youtube.com/watch?v=nxcOTVuXcUI"]Click Here For An Example Of WTF Edge Modelling Looks Like[/ame]~ Essentially, I almost always start modelling from the eyes outward, with loops around the eyes keeping roughly the same number of polygons on the top as the bottom. Although if there were more polygons on the bottom by a little, then this would usually be okay.
~ I would also make sure there is a loop running across the nose around the cheek and then through the chin. This is helps deformation when smiling or frowning as these parts all move as a family of shapes when animated.
~ A common mistake I often see is to make the eye loops just radiate outward on and on. This is a bad thing for various reasons that I'm far too lazy to explain. Anyway, to stop this I would run a loop from the top of the nose across the top of the cheekbone up around the brow and down the nose again.
~ Another result of stopping the eyeloops from radiating outward forever is a frame that goes across the forehead and down under the nose. Sometimes I do run the loop across the middle of the nose instead of under the nostrils but the theory of framing the loops is still the same in both methods.
~ I also find that if I don't have a line of polygons that run from the outer bottom eyelid towards the general direction of the nose (or under the nose in this case) then its very hard to change directions of other flow lines so I always try to make sure its there.
~ Ideally you want to have next to no triangles on a face mesh because they just don't deform well at all under animation. Changing directions of flowlines without creating triangles is a bitch to learn, but it's pretty straight forward once you understand the power of 5poles. A 5pole is where 5 edges splay out from a vert to create a group of polygons that allow change of direction. I've labelled where I usually put my 5poles. They are usually a slight hindrance to animation but not as destructive as triangles, so placing them in the correct places is very important to learn.
Above is a picture of the real mistake in not using 5poles. Faces become a flow of quads that just flow back across the side of the face without following any real facial contours or bone structure. Don't let this happen. It's bad and will cause polar bears with chainsaws for arms to break into your house and attack you.
~ Make sure all your flow lines are smooth and fairly evenly spaced. If areas get suddenly dense or sparse it will look horrific when shaded or have odd spots on it when textured, so remembering to keep the edge flow smooth is pretty important. Starting with a fairly low amount polygons and then making it denser later makes topology a lot easier to control. Unless of course you don't like control; in which case you should start with hundreds of tiny polygons and you can bathe in your own frustration and masochistic ecstasy as it becomes more than you can handle.
I'm going to ask others to continue to jump in so we can split this topic apart. I'm new to the industry and still learning too so I expect a wall of theories bouncing around!!!
Replies
Thanks dude.
This whole issue basicly only applies to SDS modelling as triangles there can totally become a problem with the topology as triangles, ngons or poles can have bad influence on the stretching of polygons with the catmull clarke algorithm.
Basicly i'd say working in quads is just more comfortable because selecting rings and loops are great, but other than that tris don't deform any different to quads as they are the same.
~ Yep. I'll agree with that. My main reason for saying keeping things untriangulated is because it can ruin a flowline if you're using it change a flow line in a weird direction.
So in that respect YES tris aren't really all that terrible especially as the whole thing becomes triangulated when it gets exported to the engine anyway. Its just cleaner and clearer if you don't use them unnecessarily.
Thanks for chiming in Neox
-#1 and 2 : two big loop around the eyes
-#3 : one tiny loop around the fleshy parts of the nose (around the top of tip and nostrils)
-#4 : one big loop right under the nose, going around the corners of the mouth with a thick margin, all the way to the front of the chin,
- #5 : one bigger loop around that, going from over the tip of the nose, hence including #3 and #4
-#6, if necessary : including the whole "monkey face", including all the previous ones.
It shows here :
http://www.yorku.ca/earmstro/journey/images/facial.gif
Once those are in, it's just a matter of flowing edges inbetween.
My take on it :
http://www.pioroberson.com/svgalleries/piorfemale/images/pior12inchfemale%20(94).jpg
BUT it's just topology discussion, and there are many different possible ways to do it, I think.
(it's more fun to make tons of cool assets than worrying about topology forever hehe)
While realtime is different from CG animation, the methods are close enough nowadays to be able to share workflows and ideas...
Poles are OK, just don't use too many. I personally prefer to avoid triangles completely, they mess up stuff in Zbrush and we always sculpt details on the final meshes, not on the concepts (bad poly distribution, lack of proper reprojection to preserve details etc)
With faces, the primary goal is deformation (facial animation) so your loops have to accommodate skin motion first. So follow the direction of wrinkles and folds with your edges. Bone structure is secondary, and sometimes you need to ignore muscles too.
Modeling from a box is old school (I've been a big proponent back then). Today it's far better to sculpt in Zbrush, get all the artistic issues decided, and then move one to retopo and concentrate on the mesh only. Faster, easier, with superior results.
Pior has some good points too.
And Yes, the thing is that there's no perfect topology, there are too many ways of doing things.
I make different topologies :S
Old school? you should say for games...
Sculpting (classical or zbrush/mudbox) and Retopo in my experience is totally not a workflow exclusive to games, its just more obvious there.
I think at the moment your mouth topology bothers me as it doesnt retain the circular topology that the muscle does and the nostrils need more refining as theres no cavity behind them and the nostrils join straight to the lips in a very flat manner, theres usually a tiny bit of cartilage there that bumps out. Its also very important to sustain the nasobial folds or laugh lines, which are the turquoise blue section. You sort of have it but it needs to go further down and that pole could be a problem by the nostril. The nostril to eye is unnecessary I think, just make sure the polys flow nice in that area as the distortion is very minimal compared to the main facial structures.
Tris for games are fine as the whole mesh is triangulated, maya users may find normals jumping on quads in places, so its not unknown to add triangles in certain places to force the triangulation on certain quads, sometimes its required aswell if the game engine is not triangulating it correctly. If your workflow is to sculpt ontop then just hide the tris best you can in areas you know do not need too much sculpting on. Keeping it quads best you can though is always nice as its not a nightmare when giving it over to another artist or in a production pipeline.
just my 2 cents
Here's an example of a production model, by Eklettica:
Just see how different is his topology.
http://www.eklettica.com/connwere1.html <--- Production model
http://www.eklettica.com/halowars.html <--- another one
For games, the workflow is totally different. I laugh when i read that with Zbrush we don't need to work in the "old way", that's pretty ignorant.
And i never have said retopo is exclusive to games. If you are not scanning a real sculpture, we say it's useless when we are working from a concept, because it would be a total waste of time and effort to work in that way. Just try to retopologize a sculpture into something as detailed as the linked models.
For game models... we don't need to make models like the above one, it's faster to have a cheap base mesh, and then, with the retopo, do the low poly.
There are too many ways of doing things.
regarding the triangle discussion: Animators love quads in games since they are able to skin quicker with complete loops / rings. At least thats what I have been told. Im not big on skinning.
I am now quad fanatic though. I use tris if they make sense. Just try to avoid them in areas that deform alot. Thats my rule of thumb.
Exactly. Even we have almost completely abandoned modeling from scratch and we're just a small animation studio. Large scale movie productions take their 3D concepts a lot further, they do lit renders and "texture" them in Photoshop in 2D to get approval. It's just so much better not to worry about technical stuff while the artist develops forms and shapes. It's also easier to build the final mesh once the forms are properly defined.
Our stuff is sometimes even more detailed and we do fine with retopo.
Yes. What I'm saying is that it's far more efficient and good fun if you better separate the tech and art parts of the workflow.
As an animator that rigs up a lot of faces (mostly using morphs and occasionally bone based rigs) then I would much rather there be some kind of organized looping structure in place for the mouth and eyes with as few poles and optimizations as possible.
The best place to stick some poles is in the corners of the nose or on the nose itself. The nose it always a crazy convergence zone and typically it doesn't move too much in a way that you're constantly having to select a nose loop and nudge it around. A little soft weighting around the corners of the nose and all is good.
At work we use almost the same topology that Blaizer posted but with about half the geometry (we do a render time turbosmooth) and no hair/fur so we optimize the scalp instead of make it more dense.
I will be selecting loops and growing/shrinking selections. Its not going to be impossible with a few poles here and there but if all you're saving is a small handful of tris then it's not worth it, in my opinion. If the loops can mimic the muscle structure then awesome!
Even if there are a few poles if they're in the right places it can actually help select things like the lower eyelid or the lower lip but most of the time it just sends the loop off in a crazy direction because the modeler didn't check the loops like an animator is going to use them.
One other thing to keep in mind is that if you're sticking to 100% all quads you can still end up with some crazy loops especially if you do things like rotate a quad to optimize things. So check your loops and rings as that is the method we will be using to select things in both skin and when making morphs.
I think its important for modelers to do some light skinning and animation so they know how to best flow their topology. Some of the biggest ah-haa! moments I've seen modelers have are when they struggle with their own topology and decide to fix it.
Here is how I make an open jaw morph its roughly the same method when skinning a jaw bone.
Typically I drop a jaw bone in there and animate it opening and closing, then apply skin, select a ring in the lips, normally near where the lips meet, grow that selection out a few rows below the nose.
Deselect the upper lip, turn on soft select.
Add/paint the jaw into the selection being sure to spread the soft part of the selection around the corners of the mouth and across the cheeks properly, this normally means painting out some influence, some of the selection and slightly spreading it up toward the nose. The advantage of this is 2 fold you get to scrub the bar and see how the mouth opens and closes, and you get a chance to adjust your selection.
I've seen people make a morph with just selection and its pretty much a one shot deal that is hard to get the selection to rotate from the proper place like a jaw bone naturally does. Now does that mean I rig up bones for the rest of the face? Oh hell no. I only use bones to make the first open jaw Ah shape after that I'll use the graphite modeling tools push/pull to sculpt in the rest of the shapes along with regular old fashioned poly modeling where the ring/loop selections come into play again. The open Ah shape is important as it gets blended into the other shapes whenever I need to open the jaw.
Mind you I'm not advocating that loops carry past the hairline, jawline or down into the neck needlessly but in the face it really helps.
Thanks everyone for arguing points. I never intended it to be a 'complete factual how-to'. Just wanted to show my method for beginners, and in the process spark some back and forth, so don't feel bad if you feel you can show me a better way of pulling something off.
Share and share alike
1. The jaw does not simply rotate. The actual movement is very complex and involves sliding and other kinds of movement too.
2. Even if it would, it slides under the skin, so the vertices definitely don't rotate. They just move inside/outside, especially as you get farther away from the chin.
It's very tricky to get right, involves more complex sculpting, but it's the only way to get right.
It does depend on the sounds you're making if the jaw slides forward or not. If you're making a OO shape then it slides forward, If you're making a AH shape it tends to stay toward the back. It even depends on the emotion the person is conveying if it should tilt side to side.
The point I was making is that with a jaw bone its easier to get the joint closer to a correct position depending on the rig or if you're making morphs it can slide around as much as you want. Not so true with sculpting or poly pushing, which are both less accurate and less forgiving.
The jaw bone is helpful because it allows for that level of play in the jaw. It tends to come out like the rubber jaws in "I Am Legend" if you sculpt it at least for me...
One of the knocks against facial animation in games is the lack of play in the jaw bone. This doesn't mean a liquid jaw it just mean the slippage you showed. If you study faces in motion there is an incredible amount of play in the jaw and when slightly exaggerated can really bring the animation to life, so I'm careful to pay attention to how the jaw moves as should anyone else doing facial animation. Ignoring it would be like ignoring UV's when making props.
With a jaw bone you can adjust the weighting to correct the pose and send the verts back where they came from. But with a soft select/move or sculpting, once you move it, they have to be nudged and noodled around and never end up in the same spot again. You could try snapping them to a copy of your original but that's just crazy talk, you're better off throwing it out and starting over.
With morphs You also have to check the animation on another head after you've loaded/reloaded the morph, its a lot more clicks than just scrubbing the bar and watching a jaw bone slide around. If you've tried to move verts back as in sculpting or poly pushing, the animation start sliding and moving in unpredictable ways. If you have a few morphs going at the same time those slightly nudged around verts starts to multiply and create ugly errors. If you want some sliding and swimming you can easily nudge things around a hair. But its better to have it happen because you put it there than have it accidentally be there and you're powerless to remove it if its a problem.
For the past 4 years I have been rigging up 8-16 characters per year and doing close up facial animation in the range of 5-20min of dialog per character. That's a crazy amount of work and I really love it, our games center around character interaction and interrogation. It's that interaction that our players love and one of the core things that keeps them buying our games. We're gearing up to start our 25th game and the company has been at this for over 10 years.
I'm sorry about hi-jacking your thread, hopefully someone will get someone will find my post(s) useful, as its slightly on subject... slightly...
But the general "jaw open" shape has to account for all the motion in that animated gif, that's how it looks like (we even have some X-ray videos somewhere).
The more important point is that simply skinning the vertices to a bone will not recreate the proper deformations. For heavily limited game rigs it's still OK and probably allows for a bunch of procedural stuff, but for realistic results you really need to account for all the motion.
I sometimes use a joint to position the chin itself, and also the teeth, but the vertices of the face have almost nothing to do with the motion of the jaw joint.
They are not following it, they are displaced by the volume of the bone and to some extent the muscles. So I usually use a primitive jaw bone object as a reference to sculpt the face.
A big eye opener is to get someone with some 2-3 days old stubble and film him from the side. From this view there's very little movement, because it's actually the skin of the throat that moves outside as the jaw bone pushes it. So what happens is actually closer to cloth simulation with an object moving underneath.
This is the really part to get right with a face rig, although it's only important for realistic characters with lots of high freq texture detail. A smooth stylized face can work with a jaw bone of course.
Also, there's a super cool Maya tool to use with blend shape sculpting. I literally couldn't live without it. The anim gifs show off the use pretty well.
http://dpk.stargrav.com/bcs/docs/data/lv2-paint-overview.html#paint-overview
Also, for rendered stuff you can use corrective shapes if vertices get moved around too much.
Say, shapeA is a smile, shapeB is the upper lip raiser. When you activate both it'd look silly, but you can sculpt another shape, shapeAB that defines how a smile with a raised upper lip should look like. Substract both A and B from it, then add a multiplyDivide node in your rig. Multiply shapeA and shapeB's intensities and connect the output to drive shapeAB to have it automatically applied, proportionally to the mix amount of the two shapes.
Gollum and the Avatar characters were all built with this method, that's why there are thousands of blendshapes in the rigs, to fix every possible combination.
Unless I miss understand what you're saying...
Maybe your workflow has you animating another parameter to move it forward and back but for me its all jaw action. I'll even lag the jaw behind if someone gestures with their head, just for some emphasis and a little squash and stretch, yes even on realistic characters it helps drive the animation home.
heh... yea that corrective falloff is what I'm talking about in max by using the skin weight paint tools to create the proper falloff. Its virtually the same thing and just as quick and easy to use. I personally don't end with whatever the jaw bone gives me, but its a hell of a lot quicker for me to skin it up really fast, and adjust the skin weights and the jaw bone, to get things into a start position as I'm making shapes.
The example on the left is what you get with a simple select and rotate in max. From there you have to nudge the points back into place and it gets ugly.
yea you can do that, but I think its better if you build your shape library in away that minimizes corrective shapes and give you shapes to use that are contributing to the animation instead of fighting errors and trying to cover things up. In other words I would rather have 4 extra shapes that add subtle motion than 4 that add in and 2-4 that are special case corrections.
Bolt, Rhino in particular was a good example of this. I read that the traditional corrective shapes for Rhino quickly got out of hand and that a new system they came up with was put into place Pose Space Deformers, basically a joint angle deformer on crack that respects outside shapes and deforms appropriately. So the lesson is, corrective shapes have an overhead and it can easily consume your shape budget if you're not careful. Also they don't fix everything...
As long as we're on the subject I also relay heavily on Transmographier, the shape mirroring and asymmetry channel are amazingly powerful and quick. Whatever you do on one side flips and or mirrors to the other as you're working. It also has some amazing center seam detection and verification to uphold its integrity. If you want to create asymmetry pump that into a separate channel and go right on working the same way but with asymmetry. It blows the doors off of anything I've used previously or that comes canned with any package.
http://www.rpmanager.com/plugins/Transmographier.htm
I can see how for a game with a lot of auto lipsync it's better to do phonemes and other controls specific for the mouth, Bioware had a presentation about KOTOR online some months ago describing their approach. So here I just basically info dump my experiences from another field - in case it might be useful info for someone.
Corrective shapes for realistic facial animation are a must IMHO. The smile and lip raiser stuff I've mentioned is a very good base example.
When you just raise the upper lip on its own, it makes a sort of M shape because the middle part is held back by the bones in the lower part of the nose. There's also some slight wrinkling of the nasolabial fold and the nostrils are dragged to the sides.
But when you use it to reveal the teeth while smiling, the upper lip should be an almost completely straight line, with very sharp corners. You also don't want to overstretch the nostrils, and the nasolabial fold (the smile wrinkle) is already fully deformed, too. So in this case you want to filter out a lot of the deformation and change the rest.
It is simply impossible to get all three states right using only 2 blendshapes, either the smile or the upper lip raise has to suffer. Unless maybe when you're working with a stylized smooth face. But with a corrective shape for the combination of the two, you can get all three states right.
Naturally there's a combinational explosion as you start adding other shapes like sneer, lip stretcher and so on. That's how you get to thousands of shapes, especially as you start to add 3 and 4-way combinations. But it also allows you complete and precise control, because you basically end up sculpting every possible state of the geometry that the animators can create.
For a movie or a complete game it's hard to avoid all this work, but in a 2-3 minute cinematic you can limit the actually used combinations to a reasonable number. And Gollum was an especially complex case because he had to be able to fully compress his face into a sad or angry state and still have full freedom to act and to speak. That's why they had 4 people working on shapes for months
Using pose space deformers for the face doesn't really make sense because this approach works in joint space. There are no joints in our characters, because eyelids work better with blendshapes too - there's all that skin folding/unfolding you just can't cover with simple transformations.
The hamster in Bolt had joints deforming it and that's what they were trying to fix. We have pose space deformation in Maya too, good stuff for elbows and shoulders, but never used them on faces.
Or you can also add a clothsimulation-like deformer to smooth out the skin after animation, I kinda recall hearing about something like that on Tron Legacy.
The advantage for blendshapes is 100% control and fast viewport interaction speed.
I haven't tried this out yet but one thing you might be able try is to "skin wrap" it to a mesh that is deforming and not changing. That way you can make major edits to the mesh but it won't require a reskin... at least in theory...
With skinning to bones, you can sometimes get around geometry edits if you go straight with envelops and don't paint or assign weights, which is typically pretty crap weighting.
What you end up doing in max is copy your mesh, make your changes, skin wrap the edited mesh to the original and convert to skin. In Maya you do pretty much the same thing but use transfer maps to transfer the skin weights. Which is why I suggested just skinwrapping the mesh you plan to edit, to an already skinned mesh. Skinwrap is based of of vert proximity instead of the vert numbering so you can make all the changes to your mesh you want and still get a pretty decent skin.
this is the base mesh we're using for just about everything in conquest. it's even the same head/mesh (minus boobs) that i've been using for this:
that said, if you feel the loops can be improved while maintaining functionality, do tell!
http://galeria.index.hu/kult/2011/01/21/digic/1902041_703eb933540a07a9b2944eaf4860ca35_xl.jpg
ME3!!!
Mirai had a different data structure as far as I know, stored everything in one model as sort of "states". So you could keep adding geometry edits and the vertices would maintain information associated with them.
But I wouldn't like to get to work with a mesh of 50-100K polygons that stores that much information too. File sizes would be truly amazing
This is the head mesh I defined for my last project, I found it really clean and useful for creating both male and female heads as well as really easy to add loops where required.
Did you miss Cryrid's post?
http://www.polycount.com/forum/showpost.php?p=1340897&postcount=39
Or do you mean something else?
Thanks..... does the same sort of topology is followed by big giants like PIXAR, DREAMWORKS
These are SUPER OLD now but getting info out of pixar is a tough chore...
And this shows some full wires every once in a while.
http://progressive.stream.aol.com/aol/us/moviefone/movies/2004/incrediblesthe_015960/incredibles_fl_dl.mov
http://graphics.pixar.com/library/HarmonicCoordinatesB/SiggraphSlides.pdf
Although I would like to try it for myself and see the differences between these two approaches on the same character, it seems it enhances the facial deformations on cartoon characters.
Jesus is that Sully? That chest topology is unforgivable! Thought by that stage they'd have mastered it...
Just a base mesh Im making to speed up personal projects, all quads except for a few triangles hidden around the ears. Im hoping I can use it as a low poly model nice and easy as its got all the major edgeloops in and is easy to pull around and insert/remove edgeloops where needed.
Ok so I am trying to make a pre-sculpt model (blocked model) and got this.
The front of the face suppose to have more topology than the bottom because its going to be hidden with hair and half of her head is going to be deleted due to polycount.
am I doing it right?
EDIT: and you guys dont create eyes? O.O