Tris vs Quads vs 5-sided The why.

[wasa]

First a little about myself. Like most here I’m a 3D hobbyist and enjoy being just that, not that I would mind working in the industry. I’ve been modeling for over a decade and why I’m interested in all aspects of 3D my main focus over the years has been the modeling.

Over the years I’ve read hundreds of tutorials, threads and watched instructional and time-lapse movies all in the hopes of becoming a better modeler. (practice is still the major key to getting faster and better) During that time I noticed a repeating pattern of showing how that instructor or individual models, but rarely the why behind it. So I’m going to try and explain a bit of it.

I hope to make this a series and will create a post for sticky if the interest is there.

Keep it all quads. You’ll hear this one a lot, but why?
Well to be honest it’s more of a learning thing, saying do it this way till you have a grasp of when and where to use other things. It’s a good practice and helps keep your geometry clean and more predictable.

Tris cause pinching. This is true and is due in large that the tri is 1/2 the size of the neighboring quads and divides into 3 quads so at least 50% more geometry for the space. This isn’t always a bad thing, and tris can make life so much easier when you learn where it’s ok to use them.

5-sided polygons. I’ve read posts ranging from absolutely do not use them to hide them in corners and under the hair. They really do get a bad wrap. Part of the reason is texture distortion (Creating UVs distorts the texture as well (just saying)), this is in large part due to the use. The other reason is they create poles with 5 edges which can create problems when animating. Even with their drawbacks they tend to work much better for changes in mesh density than tris.

Studying the wire-frames of professional artists we’ll see that main of them use tris, 5-sided, or both. They are able to do this efficiently and effectively through experience and I’m sure some trial and error.

The main things an artist should have a solid grasp of before using either of these is how model with quads. The reason being that it will help teach where one or both of these are really needed, and where a quad works fine. In the meantime try using them to build light weight control cages that will be subdivided for the final control mesh. I do this for my sculpt mesh, it’s quick and I end up with the same geometry I normally would in less than 1/2 the time.

Pre B-Mesh integration, try Wings3D or VoidWorld if your not using one of these already.

[wasa]

Feel free to ask questions about this or post suggestions for future explanations topics.

[dleri]

Here’s a simple explanation: Quads are what the user sees and they are much easier to work with. Whereas, when Blender processes your model or scene it “sees” triangles and processes them as such. N-gons: Some people prefer them because they can reduce poly count in models, but most software have a hard time working with them.

Here’s another topic that’s been done to death

[wasa]

Looks like neither of you read my post, or just made assumptions as you did.
This is a teaching thread, not a question thread.
I’m already well aware of the benefits and limitations of modeling all quads.
I’m also aware that organic surfaces are not smoothly polished car bodies.

As for that thread started by toontje, I was one of the contributors to the
original thread started by someartist over on subdivisionmodeling.

And as for N and E vs tris and n-gons they are the same thing, they interchange.
But using a pole with quads when you need a tri or ngon is silly.
I’ll post some pics tomorrow to show what I mean about the 2 being the interchangeable.

[encn]

Could you maybe show practical examples of good tri-usage, and and how it beats pure quads in that situation? I’ve been pretty quad obsessed since I started modeling, so starting to make use of other poly types is quite a hard step.

[wasa]

Ha ha, sure, and you don’t seem to be alone, too many get told quads are better, then shown examples of when you shouldn’t use the other two. I could do the same for water being unsafe to drink. (seriously, you can drink to death, and i don’t mean drown)
3D modelers seem to be so stuck in this mind set that only quads are good that they end up forcing quads where the others would work fine. The problem becomes dense meshes, or terrible topology to over come the limitations of a quads inflexibility.

I should be able to post a few examples along with the interchangeable images tomorrow.

[wasa]

Here we have the results of subdividing a few times.

As we can see the tri creates a N-pole, and the n-gon creates an E-pole.

Lets take a look at the much feared n-gon.

Hmm, not so scary, in fact that looks like a common diamond setup I see “quad only” artists use all the time.

Here we have a mix of everything next to everything to show any odd behaviors.

Yet we still get quad strips (my next thread will likely be about the misconception of edge loops)

The main thing to remember is that the flow of polygons will always follow the edge to the center.
In the case of the quad it’s only easier because you have 2 sides in parallel.

I’ll post some more examples using some of the quad only stuff from the poles and loops thread.
And still to come, some practical examples of their use.

Feel free to search google for wireframes of Stalberg’s work in the mean time. Soft smooth surfaces with tris and ngons.

[RickyBlender]

don’t forget that we should have also the B mesh hopefully soon that wil change all this i guess

and what is it going to change to E and N poles not certain !

so if you know about this B mesh that would be an interesting subject
to see what else will be able to do and what the pitfalls are !

if possible any chance from this to have a script to help with these poles on a mesh
but no certain yet how it could be done!

like able to add a T loop or C loop D loop ect.
i mean there should be a simple way to make a script to help i think!

hapy 2.5

[wasa]

@RickyBlender

don’t forget that we should have also the B mesh hopefully soon that wil change all this i guess

That is part of the reason for writing this, the other reason is I’m tired of seeing good topology getting critisized by people that quads rule all simply because someone said so.

@RickyBlender

and what is it going to change to E and N poles not certain !

If i’m understanding this incomplete sentence, number of edges 3 4 5 etc, become that many edges to center vertices.
So tri becomes N-Pole, 5sided becomes E-pole etc.

@RickyBlender

so if you know about this B mesh that would be an interesting subject
to see what else will be able to do and what the pitfalls are !

The pitfalls are pretty much the same, you get the same distortions from a tri as an N-pole.

@RickyBlender

if possible any chance from this to have a script to help with these poles on a mesh
but no certain yet how it could be done!
like able to add a T loop or C loop D loop ect.
i mean there should be a simple way to make a script to help i think!

These are more artistic than a computation, and edge loops don’t really apply well to Catmul-Clark.
(the subdivision method used in blender)
Edge loop modeling coined by Bay Raitt some 15 yrs ago applies to Doo-Sabin.
Bay Raitt used Doo-Sabin as well which suits his edge loops, no poles methods.
If you’d like to check out how Doo-Sabin behaves you can enable it in wings plugins and give it a whirl.
Wings’ Doo-Sabin implementation was developed by myself and coded by Dan Gudmundsson.
But basically if you have no poles, the model will remain no poles, but it creates tris and ngons.

Since we discussed a bit about how Doo-Sabin works, lets take a min to talk about Catmul-Clark. (CC for short from here out)
When CC divides a surface it creates a vertices at the center of each face and divides all edges in half.
Due to this nature the faces become more important than the edges. What I mean by this is you could
create one of those “edge loops” or you could create a face strip and the face strip will perform the same
basic functions. Like easy selection of the strip, allow cutting of more “edge loops”. The faces control the
resulting CC topology, not the edges. Doo-Sabin is the opposite edges are key, faces less so.

[organic]

@wasa

Lets take a look at the much feared n-gon.

Hmm, not so scary, in fact that looks like a common diamond setup I see “quad only” artists use all the time.

Someone please correct me if I’m wrong, but surely that’s not an Ngon is it? An ngon is a face made up of more than four vertices and blender doesn’t support ngons, yes?

Isn’t that five quads joined together?

[dleri]

@wasa

Looks like neither of you read my post, or just made assumptions as you did.
This is a teaching thread, not a question thread.
I’m already well aware of the benefits and limitations of modeling all quads.
I’m also aware that organic surfaces are not smoothly polished car bodies.

As for that thread started by toontje, I was one of the contributors to the
original thread started by someartist over on subdivisionmodeling.

And as for N and E vs tris and n-gons they are the same thing, they interchange.
But using a pole with quads when you need a tri or ngon is silly.
I’ll post some pics tomorrow to show what I mean about the 2 being the interchangeable.

Like I said, wasa, this is a topic that has been done to death, on here and on other CG and 3D forums.

[rcpongo]

@organic

Someone please correct me if I’m wrong, but surely that’s not an Ngon is it? An ngon is a face made up of more than four vertices and blender doesn’t support ngons, yes?

Isn’t that five quads joined together?

Correct. No N-gons in Blender,… the picture does show the different poles though.

[rcpongo]

Something else worth pointing out is that quads are made up of triangles, and this can make a difference on the model, particularly when the model is lower resolution.

As proof, create a plane object,… go into edit mode and select 2 corners that are diagonal to each other. Now move them up in Z. You will clearly see that the plane is made up of 2 triangles. There are times (when you are not subdividing later) that you need this triangulation to be the other way. Converting the face to triangles and turning the edge will show completely different looks.

This always happens with quads that are non-planer, but as the polys get smaller it becomes less of an issue. (When you add subdivisions, individual quads get smaller and more planer)

Anyways, this is one example where triangles may be preferred, but you will not be able to subdivide without bizarre results.

[HSovieticus]

So if someone wants to use triangles for the final mesh, but also wants to be able to subdivide for the purpose of creating a normal map or displacement map of some type, then what would that person do?

[RickyBlender]

well i have one example for a F gon in blender which i think is made with 5 edges but did not use this very often
in any case soon we should have the B mesh which will allow as many N gones you want
but not certain how the rendering will be affected

another example even if you make an object with mostly quads it does not guarantie that the render will be clean and nice
i did one model today for a simple slab for a capital with 6 sides and the only way i could make it look nice was to use the edgesplit modifier!
i mean sometimes you don 't really have a choice
and let say that i use edgesplit as the last tool normally but i did have to use it on many models that i did in the past!

but hope you can show us how this new B mesh modelling will work and what will change or how we will make modelling in 2.5 with B mesh tool
that will be a very interesting subject!
i did not really follow the B mesh thing in the past so i don’t know much about it but would like to learn more!

thanks
happy 2.6

[lalen]

Thanks for this thread and don’t listen to all the negativeness. If you deal with ba be ready to get negative post. It’s a way of life on ba.
I had not a clue about this stuff but am very glad to know these things and thanks again for explaining it all in detail.
Lalen

P.s. could you do one on low poly vs high poly and maybe a pm letting me know you did it so I can read it.
Thanks again
Lalen

[Shane_Plew]

Is it all right to use tris when closing a circle or is it still better to find some way to use a quad?

[wasa]

@organic

Someone please correct me if I’m wrong, but surely that’s not an Ngon is it? An ngon is a face made up of more than four vertices and blender doesn’t support ngons, yes?

Isn’t that five quads joined together?

Correct, it is made up of 4 quads, that is also the structure of an ngon when you divide it once.
I wasn’t using the BMesh build due to some crashing.

@rcpongo

Something else worth pointing out is that quads are made up of triangles, and this can make a difference on the model, particularly when the model is lower resolution.

Yes and no, what you see on the screen and in a final render yes is 2 triangles. At the same time the internal data is a quad, at least in some applications. I don’t know much yet about blenders internal data handling.

@lalen

Thanks for this thread and don’t listen to all the negativeness. If you deal with ba be ready to get negative post. It’s a way of life on ba.
I had not a clue about this stuff but am very glad to know these things and thanks again for explaining it all in detail.
Lalen

P.s. could you do one on low poly vs high poly and maybe a pm letting me know you did it so I can read it.
Thanks again
Lalen

Thanks for the words of encouragement!
Low poly VS high poly, well that’s a pretty broad topic, could you narrow it down please?

@Shane Plew

Is it all right to use tris when closing a circle or is it still better to find some way to use a quad?

I try not to exceed 4 tris in a circular pattern typically, if you’re capping a cylinder or other flat circle then it should be fine, just make sure to create a face strip on top to avoid the distortions.

Even if you kept it all quads you would still need that ring of faces to prevent distortions.

[wasa]

So here are some more images to help explain.
Please forgive the shrinkage and the distortions in the eye holes, I did a quick and dirty conversion.

First up interchanging poles and tris/ngons.

Note that the poles have become tris or ngons both highlighted in red.
The model on the right has no poles at this point, but only due to the
method I used. I combined poles tris and ngons in my own work.

Next we have both models with 1 level of Catmul-Clark applied.

The tighter geometry is do to the method and would not affect a model that was built this way.

Finally we have what the Tri/Ngon model looks like when using Doo-Sabin.

Keep in mind that using Doo-Sabin on the original mesh would produce the same results.
This is because once you apply Doo-Sabin or Catmul-Clark they produce either no poles
as as the case is with Doo-Sabin or they produce all quads as the case is with Catmul-Clark.

I’m working on a model that is 100% modeled with poles tris and quads (ngons too if I can get
BMesh running stable enough), I hope to have the modeling stage done soon. I’ll be posting it
on blendswap so everyone can check it out, examine texture distortions and animation
deformations. It’s a fairly detailed model something I would normally just sculpt and retopo so
it may take some time. On the plus side it should offer a nice set of examples, and show how
using different polygons can help keep a mesh much lighter than all quads.

[MarkJoel60]

FWIW, these videos made by a TD at Pixar discuss the use of the Pole to change directions of polyflow.

It’s kind of like a master class in modeling for facial articulation… but if you stick with it, the videos contain a wealth of practical knowledge.

Hippydrome’s Facial Articulation Theory