This is neat -- I'm reading the code and learning stuff.
And automating things like this is undeniably fun.
But since the pink layer is derived from the wall layer, I am not sure how much of an advantage this presents in a practical sense over the Photoshop-ish approach, of laying the pink over the wall with a "colour burn" blend mode, and using a slightly blurred noisy mask (or distressing the mask layer in some other way).
That could also be set up programmatically in Photoshop (JS) and probably in Krita (Python).
Can you do similar blend mode tricks with scikit-image?
https://github.com/LeonFedotov/L.A.S.E.R.-TAG-GRL/blob/maste...
https://web.archive.org/web/20180212143400/http://www.muonic...