Guitar Modeling - Part 1

by akrid

For reference. This week (Well, just today actually), I began to model my guitar, a Godin 5th Avenue Kingpin. I've whipped up a modest preview for you folks:


I'm done the body and the neck of the guitar, which are by and large the hardest parts because of the extremely subtle curves and nuances. Looking at it again, the body looks a little off somehow, but I can't put my finger on it...

I've also started in on adding the various accoutrement that a guitar requires, each providing it's own extremely different challenge. The bridge in particular was a real head-scratcher, but I eventually solved it. I had the hope of finishing all the modeling today, but it's clear to me now that it was a foolhardy goal.

It's a completely new experience for me to attempt to model with the amount of accuracy a guitar requires. If I measure a single thing wrong then the whole thing could be ruined irreversibly. I think I'm doing a pretty good job with it thus far, but the real test comes when you put the strings on. I neglected to take a few measurements that, in retrospect, I probably should have.

If there are inconsistencies between mine and the real thing, it would help me immensely if you notified me.

One thing that proved invaluable in modeling this was using a different method of subdivision - P-Subs.

P-Subs

Remember when I talked about subdivision? Remember how I said that the Catmull-Clark method is the most commonly used? Well, there's a new exciting method that has the potential to replace Catmull Clark called P-Subs, or Pixar subdivision.

Pixar Sub-D's are, obviously, a method of subdividing meshes that has been developed by Pixar. For a long time Pixar has chosen to keep this little secret entirely for themselves in order to gain an edge on the competition, but they sell the license to this technology for those who are willing pay. Maya and Modo are so far the only ones to take them up on this offer, and have incorporated them into their respective programs.

No weights (Control), Catmull-Clark, Pixar, from left to right.

My headstock weight map. The red indicates where I've weighted the edges, green is neutral.

The main reason this method is so desirable is because you can create models without the control edges that I explained previously. Instead, Psubs rely on weight maps that inform the computer on how much "Weight" any certain edge should carry. To set edge or vertex weight, you simply select the edge and use the edge weight tool, much more straight forward than adding a ton of control lines and worrying about poly-flow. Catmull-Clark SubDs can use weight maps as well, but for whatever reason it almost always leads to pretty noticeable shading errors and is very unpredictable, as shown above. Nobody uses weight maps with Catmull-Clark as far as I can tell. Pixar SubD's are not immune to the shading errors either, so even with this improvement, weight maps still cannot be used as a complete replacement of control lines.

This is pure speculation since Pixar is generally very secretive about their methods, but I imagine that this tech has done wonders in Pixar, evident from their seamless transition between soft and hard forms in their characters. It's hard to convey just exactly how revolutionary this new method is, but you have to understand that working with control lines is essentially an art in itself. Granted, you still need control lines when using psubs, but potentially much fewer then you used to.

2,4, and 6 Sub-D levels from left to right. Top is 15% weight, bottom is 40%.
But this new method isn't entirely infallible. Depending on the quality and radius that you want your edges beveled to, you need to adjust the subdivision level to an the appropriate number. When I say subdivision level, I'm referring to a parameter that increases the amount of virtualized geometry along the mesh. The 40% weight level 6 cube above is nearly 50k polygons, whereas the same effect could be achieved in 15k using traditional control lines. That's an extremely dramatic example though, there are many ways to reduce that 50k. My point is that P-subs cannot be handled carelessly.

The big thing that causes this difference is that when using Catmull-Clark and control lines, polygons tend to go exactly where they are needed. With Psubs, there are thousands of polygons being wasted on defining the completely flat surfaces of the cube, since the entire mesh is being uniformly divided.

So Psubs have the potential to increase mesh density, but also theoretically be less dense than a regular SubD model if handled extremely carefully. You still need to pick and choose between the two. Ultimately for me, it's just another welcome addition to the virtual toolbox. I thought it was a good fit for this particular project since I'm focusing on a single high quality object and I can sink as many polygons as I want into it.