11

I am new to 3D modelling and wondering if someone could break down when and when not to use subdivision surface modelling. I am thinking more in terms of hard surface stuff. Why wouldn't you always just use a subdivision surface as it looks to me like it produces nicer looking models.

I understand how they work and the way you can make edges catch light and so on. It's not an explanation of this I am after.

For example, I've watched people model realistic architectural visualisations and they won't use subdivision surfaces for the table tops or windows, even though in my mind wouldn't it create more realism to use a subdivision and have the nice edge catching light.

Or

Lot of props, building and cars for 3D renders or games and stuff seem to not always use subdivision surfaces. Is this because of poly counts or because it's easier to get good topology flow without the subdivision or maybe something else?

I realise that it's probably personal to each individual modeller and their preference, but was hoping to maybe get some peoples thoughts on this.

Cheers!

Kristian
  • 111
  • 1
  • 3
  • 1
    Using subsurf will subdivide the existing geometry allowing you to create smother curved surfaces. If you are not dealing with curves (what you call "hard surfaces") subsurfing will only result in extra processing time and will increase memory usage. On the other hand subsurfing is useful when you want to work with the minimum number of vertices that still describe the shape you are after, while making editing easier and keeping a responsive interface. –  Jul 05 '18 at 16:53

2 Answers2

11

Sub-d surfaces generate a parametric surface with "infinite" quality that are easy to adjust (shifting holding edges, changing creasing,..), but this is what you pay with:

  • polygons are subdivided everywhere, even where detail is not needed (this is not true for adaptive subdivision)

  • most of the times the surface is not perfect:

    enter image description here

    Adding detail will pull the surface closer to the control mesh altering the original shape. Either the control mesh has to be dense enough to start with, or has to be projected with Shrinkwrap modifier back to the original surface. The first requires too much geometry to comfortably work with, the second does not produce perfect result anyways, it only lessens the issue.

It's always about quality vs. time (both render and production). If you can get away with beveled mesh, that's the method to use. If the mesh is not too close to camera, you cannot tell if it has been beveled or was sub-d modeled. Also the Meshmachine addon makes adjusting bevel easy like changing creasing on sub-d mesh (sometimes you find out you would like to catch more/less light on that edge after-all).

Imho the best hard-surface technique is to use sub-d and Modo's MeshFusion to get single shape with all edges rounded from procedural booleans. This offers great flexibility and exactly accurate results. You can create complex shapes without the complex topology and without the inaccuracy. Drastic adjustments are very easy to make.

Houdini has this addon for soft booleans and Maya has the Hardmesh addon for the same workflow.

There is also a Blender addon in wip state called Bevel after boolean, it's workflow is destructive for now but looks very promising. Another approach in Blender could be to use shader utilizing the Bevel node, however this increases render time significantly.

Jaroslav Jerryno Novotny
  • 51,077
  • 7
  • 129
  • 218
  • I am new modeling too, I want to know if game modeling don't use subD and just use bool modifior, don't care about Ngons? but if I want to create a production level modeling for Advertisement or movie, I need use subD, because the final modeling has better quality? – krosshj Jul 13 '21 at 03:49
  • @HuangJie sub-d modeling and bevel modeling is used in both productions. – Jaroslav Jerryno Novotny Jul 13 '21 at 06:56
7

The main reason is that subdivision adds verts.

No matter how good your project, no matter how much time you're willing to spend on rendering, everyone has a limit. Each level of subdivision multiplies your face/vert count by 4. So three levels multiply it 64-fold, 4*4*4.

Okay, but let's say you can spare a 64-fold vert increase. Is there any reason not to subidivide everything 3 times?

First, not everything is going to need the same level of detail. Things that are focal points will need more of it. Things that are further will need less of it. Smooth, planar things will need less of it. So you might put 4 levels on some things and no levels on other things.

You could also just create that many new verts and actually sculpt some detail into those verts so that they're not just bland, procedurally-generated verts, but contribute something more to your model, things like surface texture, micro-occlusion, UV detail. After you've done that, you won't have the vert budget to afford further subdiv.

In practice, some models benefit very little or not at all from subdivision, but for many models, subdivision is a great way to improve the silhouette without much work, and you'll probably get a lot of use out of it.

Nathan
  • 24,535
  • 1
  • 22
  • 68