Myths of modelling

yukonwanderer

I'm a newbie, and have been confused on a couple points I've read on here lately.

1. Meshes do not have to be all quads. Wat?

2. Tri count doesn't matter anymore with next-generation gen, it's mostly texture size that matters now. This true?

Does anyone else have similar issues they need cleared up? Would love to have an up to date list.

Pedro Amorim

1, true. you can use whatever you want. tris, ngons, quads. use them wisely.
2, like anything in life. moderation is key. Use the budget wisely. Just because you can doesn't mean you should. Use common sense.

pior

1. Meshes do not have to be all quads. Wat?

One way to clarify this point would be for you to list the reasons why you think that they should be. This would make the source of the confusion/misconception easy to identify.

I'd venture to guess that this is caused by hearing something like "non-quads don't divide well",which depends on context, and "non-quads are not good for rigging", which highly depends on context too. People tend to take these two completely unrelated statements ... misunderstand them both ... and then cook up a random absolute rule that doesn't apply to anything

Biomag

You can read here a great explanation by Laa-Yosh -> http://forums.cgsociety.org/showthread.php?p=8172105#post8172105

kanga

At render time all polygons are reduced to triangles. For high poly models that have to be rigged and animated a sub divide operator/ algorithm/ modifier is usually applied to the mesh. These sub dividers are non destructive and can be switched on or off, making it possible to rig and animate a lower poly version of the model making the process quicker and easier. Also if deformation maps are used for surface form and detail the maps are only effective when the model is at a higher resolution. For rendering the sub divide is activated giving a higher quality appearance to the model and deformation points on the model. Tris and ngons are usually avoided because they cant be divided and often cause artifacts when rendered. For high poly models intended for animation edge nodes (5 or more edge junctions) are usually avoided because they cause surface artifacts as well. Quads for highpoly models is a general rule but a lot of artists use non quads in particular places and in certain situations to good effect.

Game meshes are not sub divided so using triangles where you need them makes modeling much faster. Using maps to represent detail means the most important considerations are polygon direction (in the case of quads), the silhouette of the model and how it deforms when animated. Most of the resolution is concentrated on parts where the edge of the model is most apparent to the viewer/player and that depends on how close it is or where it is in the scene plus what it must do (in the case of a character's face for instance). You wouldn't plaster a model full of polys just because you can. Making a model that has polys where you need them and less where you don't makes modelling, uving and rigging faster. An efficient and well planned model is much more fun to work with then a dense mesh that holds a lot of unnecessary geometry.

Cheerio

AtticusMars

Using quads has a lot of modeling benefits since they make a lot of operations faster such as selecting loops and rings of edges/verts. It's also much easier to make sense of what is going on in your mesh if the topology is quads, it's frequently more difficult to make sense of a triangulated mesh (especially for things like characters at modern polygon counts).

Some specialized tasks specifically require quads to work such as the Follow Active Quad operation in Blenders UV tools.

Generally speaking it is usually a good idea to keep a model mostly quads for these reasons. 

Just remember that more is not better, having an all quad mesh is not necessarily better than having a mesh with some triangles and n-gons.

yukonwanderer

Thanks for all the responses, that really clears it up. It's always nice to have explanations of "why" but they seem pretty hard to come by sometimes.

beefaroni

For question #1

http://polycount.com/discussion/157205/how-u-model-dem-shapes-image-ripped

tulkas09

#2 is mostly true in new/good game engines, I was always told that is something is round make it round, like it doesn't matter to have a "small" ring with 200 triangles. However we had tools for lods and other optimization. Don't try this on UE4 ...

Ahoburg

1. True. You can use whatever you want, everything will be tris in the end. All quads just make modelling/uv etc a bit more convenient, but if it hindrances your work then just use anything that will do the job.
2. Somewhat? From what I understand, the consoles/pcs are indeed powerful enough that you shouldn't worry about every excessive tri in your mesh, but there are still draw calls and such and the performance doesn't hingeson tri count alone. If we're talking about videogames, then tech artist/director can tell you exactly how much tris the project can afford for one thing or another. Or just look at tri counts of things in existing games and you'll get a relatively good idea of how much is alright.

Stuart Campbell

Question 1 seems to have been covered pretty well already so I'll leave this one out.

2. In general, I would get into the mindset of adding geometry only where it is needed. Just because the new systems can handle more doesn't mean that efficient modelling is no longer required. It still all counts, whether in rendering or memory footprint. Towards the tail end of a project when things are being optimised the last thing a project needs is a lot of inefficient models that need fixing up, or something else being dropped instead. 

So, when you add details, make sure they count. Always consider how the object will be seen in game and build accordingly. If adding a few more loops to a curve makes no difference in game then take them out. 

Also, adding too much can visually impact the model in a negative way. When textures move into the distance, the mips will reduce the high frequency noise. The same can't be said for the model, unless the LOD chain kicks in. Too much geometry in a small space will cause a lot of flickering as the polys end up too small for the pixels on screen. Antialiasing can help here but that impacts rendering performance.

I think this is a skill that doesn't get enough attention these days. A lot of students are coming into the industry creating really high poly work and are unable to simplify that efficiently. Challenging yourself to do more with less will always be a good thing.

JacqueChoi

Myth that you can get an entry level job with only untextured zbrush images.

sprunghunt

One of the myths is that Tri-count is what matters. It's not true. What matters for performance and memory is vertex count. The only way to get an accurate idea of how much memory your model takes up is by counting how many vertexes it has in game. 

Modern consoles usually run out of memory before they see any framerate impact from high vert-count models.

beefaroni

sprunghunt said:

One of the myths is that Tri-count is what matters. It's not true. What matters for performance and memory is vertex count. The only way to get an accurate idea of how much memory your model takes up is by counting how many vertexes it has in game. 

Modern consoles usually run out of memory before they see any framerate impact from high vert-count models.

I believe that it varies from engine to engine. 

sprunghunt

beefaroni said:

sprunghunt said:

One of the myths is that Tri-count is what matters. It's not true. What matters for performance and memory is vertex count. The only way to get an accurate idea of how much memory your model takes up is by counting how many vertexes it has in game. 

Modern consoles usually run out of memory before they see any framerate impact from high vert-count models.

I believe that it varies from engine to engine. 

That's why it's important to test your work in-game and on the target platform. 

kanga

Stuart Campbell said:

...

I think this is a skill that doesn't get enough attention these days. A lot of students are coming into the industry creating really high poly work and are unable to simplify that efficiently. Challenging yourself to do more with less will always be a good thing.

Yeah I think that people who started out box modelling have the advantage because loop placement was/is everything. Also lack of instruction means that the very basics of poly modelling get glossed over some how. Student will download free collapsed models and whack them into their scenes. When you see the wireframes they make your eyes water.

yukonwanderer

@kanga - what do you mean by "box modeling"?  I'm learning what I think is termed "subdivision modeling" right now (with control loops and turbosmooth, and also smoothing groups).  I haven't been able to find anything about why we would use this method, and what the alternative methods are. 

MrHobo

Sub-D modeling and box modeling are kinda the same boat. Just one is smoothed (turbosmooth/smooth preview with control loops) and one is not.
Starting with a cube and building your mesh out from it along with edge extrusion (The way I learned) used to be the standard way to make characters before zbrush became the heavy lifting tool that it is today.
Considering that he's talking from a character artist perspective, if you're doing props/environments your still modeling in many of the same ways.
I think the whole "people don't know how to make efficient topo these days" thing is bigger problem for the character art side of things currently then the prop/environment side. Not saying that higher polycounts haven't spoiled everyone a bit, just that I think you see the issue pop up more for character artist.