Comment by schobi

I tried to think about difficult ways to compute the high frequency coefficients to work from the "wrong" coefficients of the first image...

But this is clever - just smash them together. Low frequency of one image concatenated with high frequency from another. This works surprisingly well!