Topology research


hhmm don’t forget an e-book as well:D


For the face, just buy the book: Stop Staring 2nd Edition.


Yar, I’ve collected a few of them over the years (but only the ones that struck me as really helpful to my learning)…


I’m torn between grid topology and flow topology. Hippy’s methods clearly work, and I’ve been referencing them to death; but I also like the “look” of when topology follows the natural contures of the body. It captures the forms better, too. If I can find a way to combine the two I think I’d be in good shape.


oh man there are so many damn techniques here that I dunno if I should continue mine:eek:. By the way the technique I’m doing is the one I learned from Digital Tutors except without a body hope someone has time to take a look at it and comment. Thanks :slight_smile:


Have a look at those grids again - they’re not that far from flowing edge loops and in most cases follows exactly the same structure, it’s just denser :wink:

Much like Gollum and Benji Button have very similar topology.


Ok, I thought I would chime in with some wires here too for a WIP I’m trying to work through. Some caveats: I 'm posting the reference and the mesh first with no wires so people see what the mesh is looking like before they jump all over what’s wrong with the wires.


Also…I’ve seen the term “Grid Topology” mentioned here a couple of times, but it’s honestly doesn’t make a lot of sense considering a grid doesn’t seem like it would have any kind of topological structure relative to anything organic, just hills and valleys of uniform grid structure? Maybe I’m misunderstanding though what exactly is grid topology?


Compare the topology here or here to things like this or this

Basically grid topology is more uniform in it’s density where as topology is more concerned with trying to create the lowest possible resolution model that still looks good and deforms the way you need and can support the details you need. Grid tends to ignore that whole kettle of fish and go for a dense even mesh.

Realtime rendering, eg, a game engine, would never really want grid topology since deformation would be a bitch to deal with and the model file would be bulky for little benefit. This of course is a hardware limitation not a software limitation.


When I talk about grids I mean that there are large patches with all-quad topology, separated by strategically placed poles where the surface details change direction… And I consider Gollum’s face a fine example of this.
Don’t confuse it with a simple box design where detail is implied thanks to high mesh density, ie. most Zbrush models made for illustration purposes.

I do agree that it doesn’t aim for the lowest possible polygon count but that’s unwise for animation anyway.


Good points - I hadn’t thought of it like that.


I think you’re mixing modeling technique with rendering mesh strategy. The Benjamin Button head model is really nothing more than a correct, probably all quad, edgeloop model with probably two levels of Poly smooths applied. It seems like some directors/producers etc. require this for preview purposes. Even though SubDs are more efficient for rendering in a lot of cases for realtime previews and especially in Maya viewport display speed of SubD deformation just isn’t that hot.


The BB head mesh is that dense. You can’t get topology like that with subdivision, and there are geometry details there too. It also makes sense to use that many polygons as they have scanned data to drive it.


I didn’t work on BB to really know how their character pipeline works, but we must be thinking about two different things. It’s all too easy to get meshes that dense and even denser using standard poly smoothing logarithms that “smooth” and add geometry at the same time. Every 3D app has has some form of poly smoothing. It’s even relatively easy in most apps to generate a dense poly mesh from a medium poly SubD mesh and all of the wrinkle /crease topological changes accumulated in morph modeling. The topology is respected and propagates right through the smoothing operations even on the the base model with no facial expression which can then be used for MoCap sessions (in Maya and modo this can be done for sure). Sometimes details can get “Smoothed Out” and that’s just indicative of something that didn’t get modeled correctly on the lower density mesh, but all new geometry is only created between the existing geometry with minimal if any shifting to the original edgeloop structure. (a little too easy to “up rez” as many of my instructors would have pointed out back when I was in school)


You need to understand the difference between using a mesh with subdivision on top of it, and using a mesh with higher density from the start. Facial animation is highly dependent on using lots of soft falloff for the skin movements. Even a pucker (or a simple “OOO”) shape will affect the skin as far as the back of the jaw bone.

For a movie resolution character you need to have full control over the skin movement. So you want as much detail in the mesh as you can manage. For example if you have a loop defining the jawline (if it can be defined at all, like on a young strong male), then you’d want at least one paralel loop on each side to make jaw opening shapes and sneer shapes. You need to slide the loops on top of the surface to both maintain the forms of the jawbone and create the illusion of skin sliding on top of it. But one loop on each side isn’t really good enough, you’d be better off with two - which gives five loops for the jawline alone.

Then you start to apply the same for the nasolabial fold, the lips, the eyelids, eyebrows… and you pretty much end up with either Benjamin or Gollum. Now as BB relied more on mocap and scanned data (with animated displacements), and it had an almost 1:1 mapping to Brad Pitt, it was easier to go with a dense but more simple topology. Whereas Gollum is hand crafted and his loops are worn in through trial and error so it’s a bit harder to understand.

So, again, you really can’t use anything but a lot of vertices for these two:

  • proper soft falloff to distribute skin movement
  • maintaining the underlying bone structure and tissue mass

We’re just wrapping up a character heavy production and while it has taken considerable time, I’ve still been able to complete about 200-220 blendshapes (including corrective shapes) with an average poly count of 3000-3500 per head. Poly detail hasn’t been a bottleneck at all, in fact I’ll increase it a bit in the future as it’s quite manageable and will provide more control. And Benjamin’s mesh looks about as more detailed as I’d need to go if we had increase final resolution from 720p to movie 2K.

All this is true when you want to animate the face, obviously. But that is what BB was about :slight_smile:


Thanks Laa-Yosh, some really great information regarding facial animation, reminds me of the Maya Techniques Hyper-Realistic Creature Anatomy book with Erik Miller and Jeff Unay, totally understand your last explaination and the difference between the two examples.


Need to refine more near nose and mouth area…


The general proportions of the head are off; the eye’s should sit halfway down the face, the end of the nose halfway again down, the mouth halfway between the end of the nose and the bottom of the chin.


This is correctl. The more points you have, the more control you have. It will always come down to that.

It all depends on your goal. Feature film hero character = lots of points. TV commercial cartoon character = you can probably get away with a simpler subdiv cage.

It all depends on your goal. For detailed characters, an evenly spaced, dense mesh is better for more natural skin movement. It gets even more complicated if you’re talking muscle/skin sim/fake etc because edge flow may need some extra considerations. Generally I keep things flowing where key deformations take place and dont shy away from dense meshes.

There is no need to make something overly complex but there is a need to have a dense mesh thats within reason and well laid out. You will always have better deformations, better volume retention… its better for skin sliding etc. Yeah managing lots of points can be a headache so keep it realistic… keep it within the requirements of your project.

The thing i’m pondering most lately is the 2 theories behind edge flow. Being more evenly grid in U and V, or looping the larger more major masses. I tend to blend the two techniques but I’m not decided on if i like for example looping the tricep around the elbow or the abdonmen/rib cage arch, versus, keeping them more grid like.


Hey Guys, Im prototyping a base topology for majority of my human male heads.

after reading the subd thread and quite abit of this one, I am moving from C Topology to X Topology.

So if you look at my awesome attachment!
The subd version (a quickie i know the proportions are borked,) in the image is mine and I’m comparing it to an image from the other thread. I have moved the smile loop to go around the eye so I can have closed loops around the eye.

I want to know if this is a better than the original in terms of deformation and easier to control detail where the original has a spiral loop for the eye. The only difference is the smile loop is not in its normal bridging across the middle of the nose but travels around the eye first and bridging higher.

Im worred about the smile face loop because Ive hardly seen it done this way apart from a few models admittantly in the subd thread and would like some advice if I should stick with the original?

But on the other hand in the same thread some artist mentions spirals around eyes = Baaaad!