One of the toughest questions to answer sometimes is whether a particular feature of a model should be geometry or texture. I’ve found it really tough to know the answer for sure.
There are some obvious cases where a particular part of a model must be geometry or must be texture. Obvious larger shapes, for example, are always geometry. Color, or the material of a surface, are obviously texture. But there are some cases where it’s not easy to know if it should be modeled or textured. Rivets, on something like a plane, for example, do not necessarily have to be either. Rivets can easily be modeled in and can also easily be texture. There are many similar cases with almost all sorts of models, from architecture, vehicles, to plants and animals. Both texturing and modeling can be used to achieve results. Which one should be used?
The answer is, as always, somewhat complicated. There are scenarios where one method favored over the other. Much of it boils down to the art style. Realism generally will favor the usage of geometry, especially if you are creating just an image. Why? In real life, things like rivets, or panel lines, are not ‘surface detail’, nor are they part of the material. These are pieces in real life. I have seen some artists who, rather than use a texture map for bricks, have actually modeled each brick into the geometry and then added a general texture map. The results can be a lot better, depending on the rendering engine. The reason is that the rendering engine is then able to calculate the interaction of the light with all the geometry. Rendering engines cannot perform similar calculations with texture maps. This is not to say that texturing cannot produce similar and good results, just that generally, realism prefers to have more geometry than texturing. The rule of thumb here is that if it doesn’t look like surface detail (i.e. very minute details only perceivable on the surface itself), it’s usually best to model it. Other styles, such as cartoonish, etc., can potentially be indifferent to, if not favor, the usage of texturing over modeling for the converse reasons as stated before. In the case of cartoonish styles, it doesn’t really matter as much if the details are modeled or textured.
Here’s some interesting example pictures from my own projects:
This was rendered in Arnold. Here, the panel lines are almost completely texture, except for one square like panel on the main wings. The rivets are also completely textured.
This project, by contrast, is completely untextured. It’s not even finished yet, but it should serve as a good example of what tons of geometry can do. Imagine this project with highly detailed textures. The differences will show.
However, there are times when even realism will prefer textures over geometry. I may sound like I’m contradicting myself, but one of the things that we haven’t discussed is rendering engines. In particular, using any real-time rendering engine, for something like a game, will generally not prefer too much geometry. Now this isn’t to say that real times engines cannot handle lots of geometry on the screen at once. They are more than capable of handling geometry. However, there are two caveats to this. The first one is that the more geometry there is on screen, the more we must lower the draw distance, or the maximum distance from a camera at which the game engine will render something. This is done so that the frame rate of the game does not drop to unplayable levels. The second caveat is that there is an upper limit of geometry after which the engine will be unable to provide good frame rate. Sometimes these issues can be offset by using a more powerful machine. But the overall point is that for game engines and real time engines, geometry is generally more expensive to render than a simple texture map. Again, more geometry is still better for realism, but due to optimization, we have to make sacrifices.
In short, there’s never a simple answer. It always depends on what you are trying to accomplish. Realism will always prefer modeling the geometry over texturing it as a normal map or a displacement map. However, there are considerations to be made to the rendering engine being used. Also, there’s cases where it may not be feasible to actually model out geometry. Carvings, for example, are notoriously difficult to model in something like Maya. In many cases, it’s simply simpler to use textures, since the results won’t be that far off from actual geometry. For the carvings example, many artists will use ZBrush to first sculpt the carvings on the original model, then generate a texture map from the sculpt. Rendering engines are getting better, however, so perhaps in the future, we will see more polygons pushed onto the screen and a lower reliance on texturing.