Higher res and lower tris: How to stay constant(ly) (low)?

Support Modeling
1

Hello everyone,

I’ve been using Blender for some time now, but I still have one thing that I’ve not been able to quite get my head around:
When it comes to modelling higher resolutions (especially in hard surface modelling), what are the general techniques that are applied, and mostly, how does one keep their tris count (or faces etc., whatever you prefer), in check.

For example, recently I made this:

While I’m fairly content with the result, the amount of tris is rather high (nearing half a million apperantly).
The way I’ve learned to approach higher res models, especially hard surface, is to make the general shape, then apply subsurf, and add supporting loopcuts (Especially for this model, I’ve pretty much taken the same steps as was done in this CGcookie tutorial https://cgcookie.com/archive/weapon-modeling-an-m4-rifle-in-blender/).

Without the SubSurf, I have about 30k tris, still acceptable I suppose:

But once I enable it, the total goes through the roof:

Same deal with a project I was working on earlier, before the subsurf:

And after:

Again the huge increase, and that even without most supportive topology.
Now obviously this is what SubSurf is supposed to do, and I understand that, but I wonder how people can create higher quality looking models (e.g. Jisoo Kim’s great model of the similar object here https://www.artstation.com/artwork/XanNn) which sometimes just use a third of the amount of tris that my model uses without the SubSurf. I am personally “afraid” that the models I make without the SubSurf is already of too high resolution, or that I add too much uneccesary topology, and therefore results in an overall too high amount of tris.

My question is, are there was for me to achieve proper smoothing/shading with my base models without making use of SubSurf, or at least without exploding the amount of tris? I have done a fair amount of research regarding this topic, and found that some people just use smooth shading with EdgeSplit, others use HardOps addon, the loopcuts-and-SubSurf technique that I used for this, adding bevels and what not; all many ways of getting to those sharp final results, but what advice could one have for me?

If there are any rules of thumb, general approaches and/or tips that I could get regarding this that would be great.

Thank in advance! :smiley:

P.S. Here are some impressions of how the topology is for my models, if that is of any help:
Rifle:

Tracer pistol:

2

It’s not about keeping face count lower on a high resolution model but how and why are you using the high polygon count in the first place.

There are structure requirements for the model that come from the end use and the pipeline stages before that. Applications for 3D printing can only interpret real geometry so extra geometry shouldn’t hurt, until the model size, forms, and detail level makes the file size so big it becomes a problem. Animations often require deforming geometry, usually connected, and requires more control geometry but can use high polygon counts. Games don’t usually make use of what you would call a high resolution model, even though the whole term is relative to something else, but the high resolution version(s) are used to help with creating textures and fake detail to optimize the models for realtime use and still make them look good.

Silhouette of the overall model and protruding details (overlapping forms along the view) on it are important because those need to be described with real geometry. Distance to the camera is also important because it also can directly affect how much geometry is needed for making the silhouettes. Can’t dismiss that in subdivision modeling either, because curved surfaces should use same/similar density all around the object so that same subdivision level is enough to make them smooth. Otherwise another level of subdivision would make polygon count much higher because it’s per object. Each subdivision level adds 4 times more faces on each added subdivision level with quads, (quad faces) * 4^level in total.

So how to reduce polygon count on high resolution models:

  • Make it high resolution when it matters. Would be stupid to use high subdivision level for a cube, but a ball makes more sense. If the model, or one part of it has one or two curved surfaces but is otherwise angular, probably doesn’t make sense to use subdivision surfaces for the whole thing.
  • Use similar density on curved surfaces/silhouette inside one object when using subdivision surfaces
  • Higher resolution for objects closer to camera
  • Use multiple objects for different subdivision level, different object type, instancing repetitive geometry
  • As low resolution models, high resolution models could make use of fake detail like normal mapping also. Just that it’s a detailed model doesn’t mean you couldn’t or shouldn’t fake detail.

The artstation model has an example of a overlapping silhouette. The back of the grip uses less geometry on the lower portion of it so it looks angular. It’s a contrasted part of the model too so it’s really visible and could use more geometry to make it more rounded, like it is in the high resolution version.