Yeah, that’s basically what I described above with the noise layer idea. It needs to be about the same res as the heightmap used, or some multiple of that, in order to map it easily to areas on the terrain.
Really it’s just a uint8 buffer, indexed by the X/Y coordinates of the terrain squares. You just add/remove value from all three channels in order to paint more foliage down (assuming white = full density, black = nothing).
Of course, how it interfaces with zlug’s system is another matter - there has to be a translation from the color value to a spawner density in the generation.