High poly modeling question (for normal maps)

KevinCoyle

When modeling hard-edged geometry for subdivision (polygon modeling, not Sub-D) one thing that I'm still not clear on is what to do with all the extra edges you need to maintain the hard angle after smoothing? Its hard to tell by looking at other peoples wires I've managed to find because the images are too low res the edges sort of blend together. Some edges naturally flow into others but some I don't know what to do with. Also if I want to put alot of detail in one area but don't need as much edge definition in another, adding all those extra edges will make of mess of things, and make the model lumpy when smoothed.

Heres an example of what I'm talking about. In the image below I have those ridges which require lots of edges to define. Do I have to continue those edges throughout the rest of the model? Right now I have it as a seperate piece so when it's smoothed it doesn't get nasty because of terminated edges but it will likely leave artifacts in my normal map.



I've read many times that models for normal map generation should be a seamless mesh, I'm just unclear how to achieve that without ending up with an unworkably high-poly low poly cage before smoothing.

Rhinokey

i've never thought they had to be seamless,

Eric Chadwick

This is a fine way to do it, unless you want a softer transition there. Overlaps aren't a problem, usually the normalmap raycaster will just grab the first intersection it finds for a pixel, tossing out any further surfaces.

And as long as you have subsampling turned on, you won't see aliasing at mesh junctions.

Mongrelman

Models don't need to be seamless to generate normal maps from them.

What you might find easier is to use edge weighting on a subd model. You basically select some edges and alter the weighting, which will determine how sharp the edge is when the model is subdivided. So you can get sharp edges without adding any extra geometry.

Then when you want to bake the normal map (if you are exporting the model to do it in another program), you do a 'hard' subdivide, so the level 2 subd model (or however many times it was subdived) is what the actual geometry becomes. So you export this and it will look the same in the other program.

If you try exporting the base cage with edgeweights assigned, those weights will most likely be lost and the model will just be smoothed as normal.

KevinCoyle

Thats good to hear. It makes the most sense, and I haven't had problems rendering normal maps from geometry in seperate parts like this before.

I thought I remember reading on several occasions that seamless models are preferred for normal map rendering, for example, on Epic's job openning page for high res modeler they say they are looking for artists with "ability to model clean seamless models from concept art."