Ok, in that case you may end up with an iterative algorithm which moves points N steps towards darker areas. You may not get the results you’re after if you take a single big step.
I haven’t tried this, but I’d start by creating a height-map of your brightness. That means starting with a dense, rectangular nurbs surface or mesh the size of your image in world space, and moving the control-points up based on the brightness sampled from your image. This way you end up with a shape which is high where the image is bright and low where the image is dark. You can then start moving your points up- or downhill.
This moving part may be possible in Kangaroo (I’m pretty sure it is, but I don’t know how), or you can use Surface CP to get the (u,v) coordinate on the heightmap closest to each point, evaluate the normal at that location and use that information to find the most up or downhill direction. Move point small distance in that direction. Rinse and repeat.