Subdivision

#1
Subdivision

Learning how to use subdivision has been a slow process, even though my single-shot renders are simple compared to, say, animated characters’ facial expressions. In particular, overall smooth compound curves that incorporate small curved details (e.g. car body with air ducts). I am starting this thread in the Tutorials sub-forum because it’s general, not a solution to a specific problem. (Yes, I know the topic of subdivision has come up a few times before.) Subdivision is so fundamental, so powerful, but far from intuitive for 3D beginners, even if they have lots of 2D experience. I hope experienced users will add posts with related basic advice for beginners, to give them a running start. What would help are posts that are general, that can be applied to a car, human, or toaster, like example jas files, step-by-step doo-dads, animated gifs, links to previous C3D Forum subdivision posts or other 3D sites, including YouTube and Vimeo. Or how to combine subdivision with a script or another modifier. Define terms as you go, like n-gon, Iterations, Constraint. Show how any aspiring designer can invest 99 dollars and get up to speed quickly and start producing professional-quality 3D work. And have free, 24-7 access to the internationally-acclaimed Cheetah 3D Forum.

I am still struggling with the process, but in my search for online help I came across some resources that would be of value to many C3D users:

Macmonkey recently posted a link to a site promoting a box modeling primer, which looks like a good deal:
https://www.cheetah3d.com/forum/showthread.php?t=12846

The book’s author also posted an excellent PDF of a basic introduction to subdivision modeling:
(“Go With the Flow” at bottom): http://pushingpoints.com/v2/got-wires/

Basic illustration of stepping-down to decrease polys (or stepping up to add detail):
https://pixelandpoly.com/img/step-down-guide.png

Simplified guide to basic subdivision dos and don’ts:
http://www.rbandeira.com/nfos/01_gr...hp?PHPSESSID=0b2a679609e663fa3d67e2fd294cef9e

Tricky parts: Note “Edge Loop Transitions” and “Triangles and N-Gons”:
http://graphics.pixar.com/opensubdiv/docs/mod_notes.html

Zoohead’s thread — making realistic cars:
https://www.cheetah3d.com/forum/showthread.php?t=10474

What I’ve learned so far:
Use all quads (see links for triangle and n-gon exceptions)
Hide unavoidable triangles and n-gons in flat areas or on the back side (no visible “pinching” mesh distortions)
Use the fewest possible “container” edges to define subdivided surfaces.
Toggle Subdivide modifier (and Symmetry modifier) on/off in Object Browser (don’t collapse yet)
Check progress with test renders using a black and white Wireframe material
Use 4-point triangles to “step down” two rows to one (to reduce polygons and mesh complexity)
Use 4-point triangles to “step up” one row to two (to add detailed areas)
“Flow” loops, no dead ends or excessive parallel loops
Use a tight parallel edge instead of a crease for realism
Make the box model larger than the default Box so Optimize doesn’t weld distinct points
Use Blueprint and Ortho-view Bezier lines as visual guides

What I am still figuring out:
(See above links)

Search keywords:
Box model, Quad, Crease, Loop, 4-point triangle, Step-down, Flow, Catmull-Clark
 
#2
I think, one of the most important things about subd-modeling is to understand why ngons and tris can be difficult to render: You don't have control about new edges you get through subdividing. And you often get points that are connected with more than 4 edges, often called poles, spiders or spikes. The turn into ugly geometry with subdivision (but are easily changed, hidden or put on a flat surface if you are a little bit experienced after a while). And last but not least, in the end, everything is turned into tris while rendered you never see (render-tris) which is no problem with all quads in your model.

Even more important is to know where you need subdivision, where it's a bit of overkill and were you use it because you are a bit lazy (I am sometimes ;) ).

To understand subdivision there is a little exercise: Create a single poly, subdivide by 1 and make editable (or use one of the subdividing tools under polygon-tools). Watch what happens, look where the old points get positionned. The knew point Do this with a quad, with a ngon and a 3point-poly. Then start to experiment, do another subdiv etc. After a while you get a better understanding of subdivision and will know beforehand how a model will be affected through this.

All the rest, how to keep a shape of a poly or how it get's rounded, according to the edges around it, is probably written somewhere several times in all those tutorials you provide links for. It's not even much to learn.

One thing about the guy you mention as 'books author'. Didn't even have to open the link to know who you're talking about. This is William Vaughan, who doesn't know it, but who helped me a lot already many years ago when I was starting with lightwave. Meanwhile he uses mostly modo and is something for that community like Frank for cheetah. He has a vast knowledge and is a very good teacher, but there are tools he uses that are not around in Cheetah. Still, his tricks, the workflow and the generic tips (z. b. polyflow) in his books or tutorials that are not meant for a specific modeling software are probably the best I ever read (I don't know the new book, though). Especially, he tries to show you a certain way of thinking that helps you a lot if you can adapt it (if you don't have already come to the same conclusions by yourself).
 

uncle808us

Active member
#3
I have yet to read the links it will take me some time but I have one burning question.
What is a 4 point triangle?

Use 4-point triangles to “step down” two rows to one (to reduce polygons and mesh complexity)
Use 4-point triangles to “step up” one row to two (to add detailed areas)
How is this done I'm hoping it's in the links somewhere. Treasure Yum Yum.
Is it the last one in this image?
 

Attachments

#4
I have yet to read the links it will take me some time but I have one burning question.
What is a 4 point triangle?
Coincidentally, my wife was telling me about a three-sided rectangle this morning. I said, “you mean a triangle”; no definitely a three-sided rectangle. What she was talking about was a steel portal frame; two posts and a beam. The fourth side was the ground, so she didn’t count that into her three-sided rectangle.
 
#5
Sorry, Uncle, I didn't answer faster. First I didn't see your post, then I had a little time problem.

Your last triangle isn't a triangle but actually 3 triangles.

A 4point-triangle is a quad that looks like a triangle. It has an additional point (as the terminology is far from clear, somebody could mean something different, but then it's no use to 'step-down' or 'step-up').

If you look at the example, you see two triangles, both as 4 point polygons, which means it's an all-quad plane here. From left to right it changes the flow to bigger tris, from down to top you get smaller ones.

But us you see, if you just use one or the other, you'll get a tri (a real one with just 3 points) which may or may not hurt your topology. To keep an allquad-mesh you'd have to do something else to get rid of that.

Additionally, a 4th point in a tris is one of several ways to get rid of a tri. Sometimes the resulting edge does not hurt or is even desirable.

Hope this helps and maybe you'll find even a use for this methods somewhere :smile:
 

Attachments

#7
If you look at the example, you see two triangles, both as 4 point polygons, which means it's an all-quad plane here. From left to right it changes the flow to bigger tris, from down to top you get smaller ones.

But us you see, if you just use one or the other, you'll get a tri (a real one with just 3 points) which may or may not hurt your topology. To keep an allquad-mesh you'd have to do something else to get rid of that.
What is the purpose of the of the step down with the 4-point triangles?
Why not just combine the 2 tries and leave as a 6-point rectangle?
Is it simply to maintain an all-quad plane ... or are there practical benefits?

(I'm not arguing; just trying to understand.)

Thanks.
 
#8
First off, this step up and step down business is not something you should use (in my opinion) that much as it introduces almost always a new problem (i. e. a tri) or in this case two poles (spiders, stars) top left of the two 4-point-triangles and bottom right. One of them can be avoided easily (spin the 'hypothenuse' of those fake triangles, or create 3 quads out of them), but you still will end up with one left that could be something like a sore thumb in your mesh. They are no problem on more or less flat surfaces, but can be visible nudges on edges or in rounded parts of your mesh.

Of course this pole can be eliminated, too, but usually only at the cost of additional or deleted edges; always something that does influence your topology, maybe in an undesirable way. So it's often better to move the pole to a position where it doesn't hurt.

Second, there is no 'right' or 'wrong' way of working with subdivision. All those techniques and ideas are helpful, no question about that, but they are no unwritten law. If you feel comfortable with tris and ngons in your mesh, no problem with that. But the probability that they somehow do mess it up is with ngons quiet strong.

So your 6-point rectangle, while no beauty, could be a non-problem in a mesh. Still usually people try to avoid them at all costs. Don't forget, if you don't see a problem in your small test render, doesn't mean there isn't one in the big render at the end for which you maybe have to wait several hours. So my advice is too break them up into quads (and maybe sometimes tris).

To maintain an all-quad mesh (the plane here is just an easy example) IS the benefit, at least with subdivision. It's just that they are much easier to handle and they give you in your render (or in another app) less surprises. You keep the control. That's a big benefit in my opinion.

Step down is just one 4-point-triangle, also step up. There are only two of them in my example because I did show both.

But in the end, those techniques shown in all the tutorials Joel has gathered, are just tips. Their usefulness is dependent on your mesh only. That's the reason why you should try, if your time allows, to learn them, to make them your own tool-set, to understand what is the best solution for the problem you just encounter. And this means lot of exercise, till you reach the point where you don't have to think at all when you will encounter some problem area in a mesh, you just solve it. For me, for example, this means learning new things with every project.
 
#9
@Hasdrubal,

Thanks for your reply.

Thinking back to some of the more complex models I've attempted,
I would have benefited from understanding these stepping techniques.

For example, on the spark plug illustration I posted awhile back.
The geometry on the threads turned out well ...
until I started tapering the ends of the thread where they meet the core cylinder.
Then there was some ugly geometry :)
I ended up adding a lot of extra geometry to get it smooth.

@Joel,
Thanks for posting these links.
 
Last edited:
#12
William Vaughan's Pushing Points Topology Workbook.
https://gumroad.com/l/TepOz

I haven't bought it yet, but it's on my list of wants.
Thanks for the link. It looks like a really useful book that will fill in a lot of blanks for me. I’m thinking of buying it.

Update:
Just bought the book from gumroad. It also includes a bonus load of exercises (PDFs and OBJ files). Once you've worked your way through the book, the exercises are designed to test what you've learnt and to stretch you topology problem solving skills. It all looks really well thought out.

Now I just have to find the time to work through the book. I think this will be come a useful resource for years to come.
 
Last edited:

Swizl

Active member
#13
Thanks for the link. It looks like a really useful book that will fill in a lot of blanks for me. I’m thinking of buying it.

Update:
Just bought the book from gumroad. It also includes a bonus load of exercises (PDFs and OBJ files). Once you've worked your way through the book, the exercises are designed to test what you've learnt and to stretch you topology problem solving skills. It all looks really well thought out.

Now I just have to find the time to work through the book. I think this will be come a useful resource for years to come.
Awesome! I'll end up buying it soon. Thanks for the info on the included bonus stuff.

William Vaughan is an awesome teacher and super knowledgable about 3d. He mostly works with Modo, but some of his stuff would apply to any 3d app. He has a lot of tutorial videos on PixelFondue. Although they are almost all focused on working inside Modo.
 
Top