I’ve been wondering, suppose you start with a high resolution makehuman character.
Then retopo it yourself to some lower resolution matching on average that first model.
Then
have a tool analyse both meshes and create a difference bump map
so that the low poly gets an improved high poly look *
I think he means it like to have the multires modifier, create a bump map for the lower version of the shape, so once generated multires can be removed, and just the new bump map can be applied.