Perlin Noise, surface deformation NOT along the normal vector

Question

I've a question about surface deformation using Perlin noise or Simplex noise. You can see everywhere on google images how these results look like, if you calculate a 2D / 3D noise and move the surface in normal direction like the image in the following discussion on stackoverflow: https://stackoverflow.com/questions/53688083/using-analytical-perlin-noise-derivatives-on-a-3d-surface

Using one noise calculation for a vertex seems to be fast, but caves can not appear. It is like generating a height map for a surface.

But, has someone tried to use 3D Perlin/Simplex noise for each dimension? So that 3 times a 3D Noise is calculated and the vertex is moved in the dimension direction?

I think that of cause caves can be appear and it maybe looks realistic as well. I've time pressure to implement some surface distortion and I am only interested in some resulting pictures, to decide using this technique or look for another one.

So if someone has already implemented it, please post some resulting images. Thanks a lot!

Answer 1

You can do this, and the results can be interesting, but they’re pretty far from looking like realistic terrain. Here’s a plane deformed with 3D simplex noise (Perlin doesn’t look significantly better):

The issue is that there’s no volume to the surface, per se, and nothing preventing it from passing through itself. Deforming it only along the normal vector mostly avoids that. For a surface that looks like terrain, with actual caves, you’ll probably need some kind of voxel system like the one described in this GPU Gems article.