As I understand it a normal map gives tighter control than a bump map. A bump map wraps a displaced surface around the original poys. A normal map takes every little texel of surface and treats it invidually. As I understand it, though, you can't paint a normal map. Perhaps as tools get more advanced there will be ways of adding detail automatically -- perhaps starting with something as simple as a Perlin noise generator, or perhaps a way to fold in an ordinary height map. In Doom3 they built real geometry to generate the normal map, then applied that to the low-poly runtime models. I could see this as a help even for those of us doing stills; it would let you model lavishly, even wastefully, then just before you move to the next object in the scene pull normals and swap out to a low-poly model.