Question about hard surface modeling

[dustinbrown]

This is something I've been wondering a lot about lately. What is the general oppinion on terminating edges in areas where a hi fidelity area meets with a low fidelity area?

When I'm modeling a gun or some other hard surface, is it perfectly acceptable to have it comprised of multiple objects in places where connecting them geometrically would significantly increase the poly count, or should I always make it a single mesh? Like lets say you have ten or more cylindrical shapes that need to jut out of a rectangular shape. Do you model them to connect and chamfer the connection points, or do you just leave them as separate pieces for the sake of conservation?

Here for example, he has a rail on top of his gun and he terminates all those edges instead of carrying them through the body of the gun. Looks great, but it creates a pretty nasty ngon.

Thanks in advance.

[Cyrael]

I feel its the general concensus that it is fine to make several objects when creating these hard poly meshes, I could be wrong, but that is what I have gathered from scouring the forums.. This is how I've done it, but I am by no means an expert, so anyone feel free to correct me if I have gotten this wrong.

[Vincent]

Hey,
I prefer to use multiple sub-objects, it's way easier and faster to work instead of having a single mesh (like you said, it will increase your vertex count, and having to subdivise a whole model just to add one or two details is pretty epic) . I guess 90% of the people you will ask will answer the same thing. The only (minor) problem would be when you will be baking your ambiant and normals maps, but you just have to render each object one at a time and you'll have no problems, maybe just one or two minor tweaks on both normals and ambiant maps.
Hope that helps !

[dustinbrown]

Awesome, thanks guys.

[MRico]

Yup, it's perfectly fine to use floating geometry.

I try not to do it as much on the high poly just for fun/practice and to get those smooth highlights on the normal maps. You wont get those if you just insert a cylinder into a flat surface. If you model the cylinder into the flat surface the high poly model is going to create a nice highlight where they connect.

On that gun you posted...I don't think he "terminates" those edges on the rail. It looks like the whole rail is one piece, just layed on top of the gun.

Like those cylindrical bits and switches on the side of the gun. They're not extruded from the gun mesh. Rather...modeled as a separate piece and have it sit flush on the side of the gun.

Edit: Yeah on the low poly don't be afraid to use separate pieces. You'll waste tris if you try to connect cylinders and stuff like that together. Just have then intersect. It's perfectly fine.

[Aftermath]

Moved to proper location.

[Beartastic]

There are two main benefits to keeping it all one mesh.

1. You intend to skin and animate it. Not that important for most weapons and trucks and hard stuff that doesn't bend. Skinned, floating geo looks terrible when you animate it because stuff that should look attached slides against each other.

2. When polys are all connected, one UV island and one smoothing group, you only need to use one draw call to render the whole mesh. This is programmer speak for "it renders way faster". It's crazy to try to optimise a mesh to only be rendered in a single draw call, but repeating pieces like the gun rail (where there are dozens of potentially floating bits) would be good candidates for attaching together.

These both apply to game-res only. For the highpoly, I say do whatever looks good. So long as the normal map looks ace, it doesn't matter how you got there.

[Maph]

2. When polys are all connected, one UV island and one smoothing group, you only need to use one draw call to render the whole mesh. This is programmer speak for "it renders way faster". It's crazy to try to optimise a mesh to only be rendered in a single draw call, but repeating pieces like the gun rail (where there are dozens of potentially floating bits) would be good candidates for attaching together.

Couldn't have said it better myself.
Although the texture coordinate (#UV islands, #UV sets, etc...) and normal info has more to do with the final vertex stream size if I recall correctly. Thus affecting general performance rather then increasing the number of draw calls.
Correct me if I'm wrong here tough.


As far as the high res goes, I personally mix both techniques. Stuff that needs to obviously look 'welded' together (larger details), I'll build in one single mesh, the rest in floating geo.
For smaller details, I sometimes just smooth out the normals in photoshop.

[Beartastic]

Couldn't have said it better myself.
Although the texture coordinate (#UV islands, #UV sets, etc...) and normal info has more to do with the final vertex stream size if I recall correctly. Thus affecting general performance rather then increasing the number of draw calls.
Correct me if I'm wrong here tough.

I think you're completely right

[dustinbrown]

I'm necrothreading this because I just came across a little tutorial that deals with this exact scenario, and I figured it would be a good thing to include here.

http://racer445.com/pages/tutorials/...r-fps-guns.php