Retopology for games


#1

I wonder how to make a pretty good retopo for games, and I see that peolple use a lot of triangulation for optimization, but handicrafted in specific and proper areas.
Take a look at this image in the link for instance. How can one achieve this result? What’s the workflow to find where and how to build those tris?
https://www.artstation.com/artwork/VdBkl8


#2

There’s no real magic formula. Just bear in mind the following.

  1. Respect the flow
  2. Protect the silhouette
  3. Mind the poly count
  4. Focus on what can/will be seen

That’s about it.

Creating good edge flow means cleaner deformation and less poly waste. Protecting the silhouette effectively means that you’re going to do your best to assign detail where it’s needed most and preserve the volume. Keeping an eye on the stats prevents you from going over budget. Focusing on and identifying the visible elements helps to keep you from wasting precious polys on stuff that will never be seen.

There IS one special case though. Toon shaded models. The retopo is almost the same, but you tend to add in extra edges to emphasize the areas that you most need to get picked up by the line drawing algorithm. Otherwise, those key details will likely get ignored or not get emphasized enough.

Beyond that, it all boils down to practice and common sense. Retopology becomes second nature after a while. Meditative, but mechanical.


#3

It won’t be subdivided, so who cares? Topology for games is quite different than for sub-d animated models. He simply ends loops with triangles, where ending with quads would make more work.
He first built main flow, and then added additional loops where needed to add more details.


#4

Respectfully, this is so amazingly wrong. Let me explain why.

First off, the assertion that, because they use more triangles and aren’t subdivided, topology somehow matters less in game models is completely false. Every model, whether it’s quads or ngons, ends up getting triangulated into strips or fans before eventual rasterization anyway. This is as true of film CG as it is game CG.

Second, full form triangulation & optimization typically don’t occur until later in the process. As a game artist, you’re going to work with a mix of quads and triangles throughout. As you well know, from modeling to UV, saving triangulation for later on just makes your life a whole lot easier purely on a visual inspection level. You can hold off until a point too since rigging is a vertex level operation, not face level.

Third, good topology is crucial when it comes to clean deformation and creating a model that gives you more detail with fewer polys. If that weren’t the case, we’d work with either rigid grids of polys or random jumbles. Neither one of those cases ends up well for the animator or the end user.

Fourth, that source sculpt is going got get repurposed multiple times over. The cinematic, real-time, marketing, and various LOD versions are going to all share the same primary source. Even if the real-time version ends up being more purpose focused and optimized, the starting point is going to be the same. The core topology that informs and defines the higher detail version is going to trickle down. Creating these other lower versions is MUCH easier to do if you start with a better developed and organized foundation. Work smarter, not harder.

Fifth, looped topology breaks down into ordered strips and fans naturally. That, in turn, leads to much better overall performance without hand optimization. I’ve been programming for 37 years or so. I can tell you that it’s FAR more efficient to write out 1 strip of 10k triangles than 10k individual ones. You’re looking at 1 call VS 10k. At the same time, even if you’re working in triangles only, you’re going to fare much better with organization. Take the example of a cone.

Ultimately, game models may be more optimized and see eventual triangulation, but topology and the process of retopology matter. A LOT. A model with good topology is faster, deforms better, is easier to work with, and can be more easily repurposed to multiple platforms and mediums.

IMO, just because you may see a game model that looks like a jumble of polys, don’t assume that it was never retopologized. It almost certainly was. If it’s ugly then that’s a surefire indicator that it was probably also crunched down at some point to match a target platform, frame rate, or a specific purpose.

FWIW, regarding that “purpose” issue, let’s also not forget to differentiate between a character model and a prop. Not all of the reasons why (re)topology is so important to characters hold true to a static, non-deforming prop. (Some do, but you have fewer considerations when you’re dealing with a car VS a hero model.)


#5

You are totally correct, I apologise for expressing myself vaguely. By saying “who cares” I meant that the notion of “no-triangles” model stems from a sub-d modeling, and is less meaningful for game modeling. One of the reasons is that game models use normal mapping for normals approximation for further surface representation.
Also, character models rarely require very smooth class-a surfaces, contrary to cars or any kind of product modeling, where a very precise modeling is required for immaculate surface representation. Characters are mostly done from many pieces, with skin shaders, some minor imperefections.
So there are many scenarios for topology:
-characters (done mainly with sub-d modeling, often consisting of many parts) A good defornation is priority, so no long triangles in areas of heavy deformation. Bear in mind, that there are many areas in a character, where the deformation is very minor. For example boots).
-game characters (is not subdivided, what you see is what you get with normal maps). Defornation is crucial, as it won’t be subdivided, and any lack of polys will hit back even harder. Parts are sometines connected over with retopology to save polys.
For a game model I would avoid triangle on a character face for sure. I would avoid them on big smooth areas, where they can be easily eliminated. But if you use triangles in areas of deformation, try to make them closer to a square rather than to an elongated rectangle.
-product modeling - a surface representation is crucial, so any triangle will be a problem where a smooth surface is needed. Leads to using NURBs or some kind of Mesh Fusion modeling.
So, in my opinion, triangles are ok. But one should avoid them when possible, but also not be crazy about eliminating them in game models.
Some crucial areas where to avoid long triangles: areas of heavy deformation. Specifically shoulders, hips area, pelvis.
I would really suggest learning basic skinning just to see yourself, what happens when the model changes its position. No amount of tutors about topology will teach you that good as several hours of posing your own model with basic bones. It will take you several weeks to get the grasp on, but will give you invaluable expertise.
Usually topology research does not take into account those differences in modeling application. It deals with sub-d modeling, which is mostly about viz (not realtime) models. And product viz rarely gets animation deformation.
When thinking of it, I realise it’s acomplicated subject explaining for someone who is just starting, and would require maybe an hour of lecture. Hence the confusion.


#6

Thx guys! I really appreciate it. I already build pretty good movie characters, but I was still clueless about game ones.
Honestly I’m trying to develop cel-shading characters and teaming up with a programmer, in the attempts to find a code that could pair the light source with a 90º angle of an object’s surface and a code to build a module of a plugin to control and deform the normals of the vertices in many different aspects.
My plan is to make rendered characters as good as the ones from Dragon Ball FighjterZ, BlazBlue and Guilty Gears, and I’m deeply reserching for it.
Hopefuly, one of these days I will be sucessful in these attempts. :smile: