Some programs have code to generate them. It's really not much of a big deal. In some cases, they have the data encoded as constants. I have icosahedron hardcoded in a array and cube in a code snippet.
[quote name='jonarmstrong' timestamp='1323685809' post='4893060']No one creates 3D polyhedrons at run time via code? Do I just need to use a 3D program to create the models for the dice?
So I would recommend looking into opengl, since that's a little quicker to learn...With C#? I'm not sold on that. I have used OpenGL for a while... to be completely honest, I suggest against using OpenGL in the first place. I still cannot understand how you people can say OpenGL is easier to learn. My main problems in understanding Direct3D9 were indeed caused by OpenGL over-reliance. Direct3D10 is totally awesome. Don't get me started on the documentation.
[/quote]
Actually most quad, box and sphere geometry used for debugging purposes is still all generated at runtime and never loaded from any file. It is the simple concave shapes that are nearly always generated on the fly, as you need these for more then just graphics, physics can make use of these as well to do collision testing.