Tutorial : "Making of Moff" by D.Pattenden, San Francisco
Subdivision Modelling
The heart of the concept is that we are no longer modeling at the vertex level, but instead modeling a ‘control cage’ that drives a higher resolution subdivided result. Here, the cube is the control cage that drives the subdivided version of it within. It’s the outer cage that we’re manipulating, not the subdivided result:
The biggest hurdle to overcome starting out with subdivision modeling, is how to control the output, since its not really a ‘what you see is what you get’ technique. There’s an element of prediction of the outcome involved which, with practice becomes second nature. You gradually begin to learn to predict how the smooth algorithm will affect the geometry. To explain, smooth a cube. Go to polygons>smooth. The settings should generally be set to subdivision levels1 ( or 2 or even more for a final render ), continuity 1, and smooth UV's on. However, the result is simply an uncontrolled blob:
However, introduce extra edge rows close to the edges of that cube, and the smooth algorithm simply has no choice but to obey their existence:
|
It’s a hard concept to explain, so the best thing to do is to play with it. Some people like to use the ‘smooth proxy’ function instead of using smooth, which is the exact same as smooth, but it will automatically place your control cage and subdivided result on different layers so that you can quickly view the smoothed and unsmoothed versions. There are in fact many different tools and setups that people use in subdivision modeling workflow to distinguish between their control cage and its smoothed output. In Maya, the most obvious are simply 'smooth and undo' ( speaks for itself. quick and dirty ) and 'smooth proxy'. But there are some great additional scripts such as smooth proxy side by side, CPS and I'm sure many more. It really comes down to personal preference.
There are often no strict rules in modeling, but loosely, these are some general requirements for successful subdivision type modeling:
1) The control cage should be where possible left as quads. The smooth algorithm simply works better with quads. It's a mathematical fact that the smooth algorithm will always turn a model into all quads, but the more quads it has to deal with than triangles, the smoother the surface tends to be. If a control cage is 95% quads, then that’s doing OK. But more is better.
For example, If I smooth this triangulated cube, my topology becomes a mess ( and therefore how light bounces off the model will be equally messy ) :
But note that If I smooth this cube that is all quads, my outputted topology is much cleaner:
|
2) Try and keep the control cage as light as possible. And similarly, don't let your mesh get too dense too quickly. Make sure you are fairly sure of your basic structure before committing to finer detail.
3) Avoid long narrow polygons.
4) Once you have subdivided your model, NEVER delete your history. You always need to retain a version of your model in ‘low res control cage’ form. Once you have subdivided, there are now way too many vertices to edit the mesh at the vertex level, rendering that mesh practically un-editable.
5) The principles of ‘edge loop’ modeling need to be grasped. Particularly with modeling faces. Very roughly speaking, edge loops represent muscle groups, but moreover, they represent the sub forms of the model that can be manipulated to most quickly define the overall silhouette and contours of the model. Refer to Bay Raitts page ( of Weta digital and Gollum fame ) to see an overview and far better explanation of ‘edge loops' here
It's worth noting that I built this geometry much before writing this tutorial, and when I look back on it I really don't like the look of the abrupt termination of some edge rows in the middle of his forehead there. Whether or not it is the 'done thing' to leave n-sided polygons in a mesh thats intended for smoothing is an ongoing debate in CG modeling. There's a huge thread on body mesh topology on cgtalk here. The thread starter was none other than Stephen Stahlberg, who sits firmly in the camp of it being OK to leave non quad polygons in the base mesh.
How you actually go about physically modeling is entirely up to you, and everybody I know seems to have a different way. Very generally though, whether you're using Maya, XSI, Silo, Wings or whatever, the techniques and tools are the same. I tend to start with a cube. I then delete one side of the faces and instance mirror the geometry. That way Im working on one side and the other is being updated instantaneously. I tend to model 'around the eye' since the eye is just the most obvious place to start to me. I then simply draw topology onto the cube, tweak, and repeat. There are really very few tools that I use for actual mesh editing. Some kind of tool that enables drawing on the mesh ( in Maya it's the split poly tool ) , extrude, bevel is really about all you need at the most basic level.
At the time I did the base for Cushings head, I did a time-lapse since it was for a speed modeling challenge. The video is about 12 megabytes. You can click to watch or right click>save target as, and save the file. The video takes the model from cube to rough head shape ( about 30% done ) Download Video
Now again, I created this model quite some time ago, so there are probably quite a few things that are much more efficient in my workflow now. I'm looking forward to lots of e-mails saying 'urgh, I can't believe you modeled it that way!' :)
If you don't want to d/load the 12 meg, here are some interim stages of the model during creation:
The ears disappear at one point, because I bolted on new ones from an existing model of mine. And why the hell not?! Ears are a bitch! Once the model reaches a certain point, it's really just a case of implicit observation of the subject to get it to where you need it to be. Try and break up the face structure as you look at it into simpler planes. Here, I've applied a cutout filter in photoshop, and painted into it a little, to try and figure out the most basic facial planes:
Once you're happy with your geometry, combine both pieces and stitch up the middle seam. You can then add some symmetry. Tarkin has a slightly broken nose for instance. Whether you intend to animate or no, it's good practice to model the face in a fairly default pose, and then model a blendshape ( morph target ) as your target expression. Edit > duplicate ( making sure you are not instancing ) and tweak away. You can use the Maya 6.0 soft selection tool for morph target manipulation. It's in the transformation tools section of the menu. Modify>transformation tools>soft modification tools. I actually prefer the awesome magnet script.
Once you're happy with your target, pull down the animation menu, select your base mesh and then your target, and go to deform>create blendshape. You can now turn your expression on and off at will.
Default pose and somewhat disgruntled expression at right:
The eyeball model is pretty simple. It's really just a question of replicating the actual structure of an eye. That way the way that light bounces off it later will be more convincing. Note how the eye has a concave iris, an actual hole for a pupil, and an outer sphere that acts as cornea with a convex lens. Note also that the geometry is all quads in anticipation of smoothing:
Final smoothed model with blendshape turned on at left and unsmoothed wireframe at right. Note from the wire that I'm less concerned about the occasional n-sided polygon in my tunic than I am the face. You generally don't want too many n-sided polygons near the facial features that might be animated:
|
