How do I get "Set Matte" to work with camera movement tracked in Mocha?

Hello, I have a composition of a person standing in front of a very bright window. I want clouds to be visible in the window, so I added a new clip with clouds on top of the main clip, tracked the window with Mocha and linked the clouds clip to the Mocha output, everything works well until this point.

Then I applied "Set matte" to the clouds clip and set its mode to Luminance, so that it doesn't cover the person in the clip, only the bright window. It looks good, but once the camera starts moving up, the matte (transparency) of the clouds clip doesn't follow the main composition, it has a vertical offset. How do I fix it?

I tried making a copy of the main clip and linking it to the mocha tracking output and using it a matte for the clouds clip instead, it didn't work.