Some thoughts on a NURBS to Poly workflow with MoI

 From:  Keris
3196.3 
Now, while the last mesh was serviceable, there are some points to note about it. For one, if you were playing along at home, you’d notice that dialing around that angle would cause the way the round handle gets divided up to become strangely shaped.




Not only does the wheel look a bit strange, the odd divisions wreak havoc upon the way the arms flow into them. To avoid this kind of mess, we need to trim the mesh so that we can force edges where we want them.

This trimming process actually requires some forethought about where to put them and how dense you want the final polygon mesh to be. For now, I’m just going to hit the high points and go into more special cases in a bit.

The general rule of thumb I go by is “don’t have anything large than a 90 degree turn about the object.” That might sound complicated, but it just means that, even if it’s a smooth cylinder, I’ll divide it into at least four sections. This works in both the directions across a complex surface.

The reason for this really comes down to the way people see things. The human brain likes pairs, triplets, and quarters. It loves these simple sets because they’re fast to process. Multiples of them are also good because the brain can divide them out decently well. Fives are good only in the case that you’re using them as part of a decimal counting system, like having five items. Things divided into fifths are harder to parse. Unless you have other, more visually appealing items. Also, for large objects, at just about any time, one can always see a quarter of the object, thus nailing down the edges helps to control the way the form is seen.
So, to divide up this model, I look it over. It has four arms, so the outer wheel will be divided into four parts at the midpoint of the arms. Had the wheel had five arms, I’d divide it into five parts with the center at the arms because the arms dictate the way the eye sees the object more than the total sides. Each arm, the wheel, and the center will get divided into four parts around the object.



I can accomplish all of this with just three lines and a circle (hard to see there …). Draw the lines out over the parts, then just make sure you move them away from the object. Next, take the wheel solid and Separate it into surfaces. Now just Boolean Difference them. If things don’t work as you expect, you might need to extrude them without caps to Boolean right. It might not Join up into a Solid anymore, so you should only do this once you’re ready to output, and probably only to a copy of the object and not the original.



With the exact same settings as before, we now have a mesh with far more predictable points of flow. Now, the arms have a crapton more edges, but those aren’t too tough to remove later (sadly, “avoid smaller than” doesn’t seem to affect them at all).

At this point, I can now actually think about baking in the number of edges I want. This kind of goes into the thought about how many edges you need to properly show a round object. My thoughts are you need to think in multiples of four. The larger the object, the more you need. For this wheel, the center I want to have at least eight in the low poly, while the outer wheel needs at least twelve. I personally think six sides looks off, and going to a low odd number also looks weird. And four just isn’t enough for something you can see the end of, no matter how far you turn the smoothing angles up to.

With those thoughts in mind, I actually decided to export out an object with twice that much density. The reason was that, based upon the way the mesh looks at those settings in the MoI mesher, there was just too much distortion where the arms meet up on both sides.



By doubling the amount of sides (and a lot of toying around with settings), it looks far better.



Now, this is where I mention the next important trick. If you end up in this kind of situation where you want some parts to be at one amount of an angle and others to be at a different one, the best method is to export the various parts separately with their own settings. It’s faster, ends up with less odd things (like those T-junctions on the outer ring). I’m not doing that here because it’s harder to illustrate the whole object without composting images together.