Tris x Quads

[madeirandre]

:) Hi, hello everyone. I'm new here and new in the art of 3D modeling and a I have many questions about when to use quads or tris. I watched some video classes and readed some books about 3D modeling for games and in most cases the guys said "hey use quads, they are better...".

But after I descovered this amazing society of 3D artisans, I saw many works of guys which alredy works in the game industry and their models were completely in tris. What's the true? When must I use quads and can I use tris?

:cry: Help me guys, please. And sorry my English.

[Swizzle]

If you're making a model that will be changed to subdivision surfaces or will be imported into something like Zbrush or Mudbox, quads are best because of how subdivision algorithms work. Using triangles or n-gons will give your models weird bumps and areas that look pinched.

However, you'll want to use triangles for game meshes because most engines (and all video cards, I believe) only process triangles. Since you won't be subdividing a game mesh, it's perfectly acceptable and actually encouraged that you use triangles.

[Aftermath]

quads are what most people work in but when it comes to games in the end when the model hits the engine its all tris anyways. I suggest working in tris so that you can import into zbrush and mess around more with your character.

[madeirandre]

Thanks guys. I don't want abuse of your generosity, but where I can find more information about how card games and engines process triangles? And maybe some tips about how to model a good character for game.

[Deto]

For character modelling I'd take a look at Athey's character tutorial which is pretty much the best I've seen so far:
http://www.bakaneko.com/howto/computer/3d/character/page01.html

Or look at Gnomon DVDs. There's some very good video tutorials if you're not low on cash :)

[YdoUwant2know]

For the most part, Poly vs tri all depends on what you are doing with the model. If you are building the model to be brought into ZBrush or Mudbox then quads all the way. If the model will just be put strait into the game, then optimize it how ever you want. for the most part, when I model, I try to use mostly quads (it is much easier when it comes to creating proper edge loops and such) but if I need a tri here or there, I will use it. As for triangulation in a game engine, most engines automatically convert everything to tri when you import so I generally don't worry about it. I am aware of how my model will look as tri, don't get me wrong, i just keep things in quads till the end.

[artician]

Slightly different question - I've never bought into the quads-for-sculpting thing. I know it's better for sub dividing on import, but triangles offer so many more opportunities for optimization that I've always just stuck with those and ran a single meshsmooth over the model before throwing it out to mudbox/zbrush.
That keeps the models shape and converts everything into quads for you. I've never had a problem working in this method. It would be nice to know if anyone else did this because literally every other artist I know just immediately switched to the other method because "that's the way it's supposed to be done".

[Bokaja]

I also started more and less ignoring a perfect quad-mesh for zbrush - it killed my inner child ((OO)) - so I just go nuts and keep my creative focus on the models porportions and silhuets. I try keep as much quad as I can but I no longer hesitate in putting in tris where I can :) A turbosmooth iteration 1 on top will fix all the tris anyways - then into Zbrush! Horayh!

[artician]

I'm glad I'm not the only one. As game artists we already walk a very tight rope between art and tech. The quad limitation, while not completely damning, was too much of a harness for me. We should be able to focus on the art more in that way.

[cookepuss]

The quad limitation, while not completely damning, was too much of a harness for me. We should be able to focus on the art more in that way.
Goodness. You all talk about quad topology like it's the bane of your existence. :) I'm not saying that it's entirely impossible to avoid triangles, but there are usually predictable places and ways to hide them.

Creating a quad mesh isn't hard. It just requires a different use of the tools and way of looking at your mesh. The same goes for triangulated meshes. You use a couple of different tools and techniques with each method, but once you're used to it the issue of difficulty is pretty moot.

I suggest working in tris so that you can import into zbrush and mess around more with your character.
Simply bringing a triangulated mesh/mess into ZB isn't the solution either, especially if your core loop topology is a mess too.

First of all, SDS favors quads. SDS can generate some very nasty dimpling, creasing, and warping in a triangulated mesh. Subdivide a quad cube. Subdivide a triangulated cube. Totally different beasts. Your intended silhouette and topology more or less get thrown to the wolves and then into the crapper.

Second, SDS turns triangles into quads. Why not start there anyway? Lazy artist. No cookie for you. :)

Third, SDSed triangles result in more geometry.

Test:
A) Subdivide a quad plane to 1 iteration. How many new quads do you get? Answer? 4
B) Triangulate a quad plane. Subdivide that quad to 1 iteration. How many new quads do you get? Answer? 6

Not bad so far, right? Let's keep going. :)

Subdivide each test case to the 8th iteration. How many total quads do you end up with for each?
TEST CASE A.... 212,144
TEST CASE B.... 393,216

Let's get more practical. How about cubes?

A (quad cube)
>> lvl0 -> 6 quads
>> lvl8 -> 393,216 quads
B (triangulated cube)
>> lvl0 -> 12 triangles
>> lvl8 -> 589,824 quads

Much bigger difference now. Subdivide each case once more, lvl9. A goes up to about 1.5mil quads. B goes up to 2.3mil quads. At this rate, the triangulated mesh will eat your RAM quicker. Never forget about the winged edge data structure that guides SDS.

In the context of SDS and ZBrush, how is triangulation good then? In my mind, that's just bad. If you work incrementally, sculpting Lvl1 first, then Lvl2, then lvl3, and so forth... You're just not getting the sort of lower level control that you might need when starting from a triangulated mesh.

Worse yet, because of the dimpling and creasing artifacts resulting from subdivided triangle meshes, you may end up cleaning up messes instead of sculpting. The time you spend on cleaning up artifacts could have been better spent getting the topology right the first go around.

You'll never convince me that quads are not better. :) They just are.

You could work with a slightly less clean/formal base mesh early on if you positively intend on retopologizing later. However, you have to be clear that the goals for the final mesh remain the same.

1) Keep it clean to re-edit.
2) Keep it clean to "read."
3) Keep it clean to animate from.
4) Get more from fewer polys.

With #4, triangles become most important in the optimization phase. Early on, during pre-sculpt and sculpt phases, quads will always serve you better.

Just food for thought.

[Bbox85]

Wow, it seems like everyone has thier own opinion on the whole Tris vs Quads thing. Y'know... I've come to accept the fact that we can't fully eliminate triangles from a mesh. It's near impossible. However, I always believe that the fewer, the better. Keeping everything in quads and having the edges on the inside flowing in one direction continuously is optimal for engines. I know that the PS2 took adavantage of a little thing called PolyStripping... atleast that's what I believe it was called... anyway, in a nut shell... you can have a large row of quads and as long as thier hidden edges where flowing in the same direction (i.e. the [Turn] button in the Face Subobject tool bar in Max), the engine would count it as one poly. The moment that contiguous flow breaks, it'll start calculating other polys. Now, I hear that the 360 doesn't do polyStirpping but I'm not sure if that's entirely true.. and the PS3 might. Anyway, I say that if you can help it.. keep it quads. When you break it down... a Quad is two Tris anyway...

[seven]

The point where I feel quads serve me the best is when I am making levels of detail. In most packages it is easier to select an edge loop and just start deleting. The process is sped up and is far more efficient than a mesh that has tons of triangles. Towards the end of my lod and optimization phase the model will become more or less quads and triangles. But the triangles are very strategically placed to define silhouette value and to make decisions on where the quad gets split vs the computer splitting it based on the most efficient tristrip tessellation value. Often this does conflict with aesthetics. Just another good reason to strategically place tris but again, prior to the optimization/LOD phase, the model is almost always all quads.

[jpgourley]

The only problem with not triangulating a model in Maya or Max and then putting it into an engine that automatically triangulates is that it can be very unpredictable. You may very well end up with tri's that are backfacing or flickering because of how the engine subdivides your mesh. I like to keep things in quads for the most part, but on a face that is drastically bent or tweaked I would definitely triangulate it. After a bit of experience you should be able to spot most problem faces before you ever export your file. You'll sleep better at night knowing that you won't be fixing bugs down the road.

Edit: On your low-poly game ready mesh only, this doesn't apply to anything bound for Zbrush or Mudbox.

[artician]

cookepuss - I wasn't saying tri's were better either, I've just met a lot of artists who have been really adamant about 100% quad-modeling being necessary, when it's not. It's all just different methods each of us have for going about it, and in the end the only thing that matters is the model that goes into the game engine.
I personally get more efficient models by using triangles wisely, but maybe that's just my limited experience. <shrug>