Thinking out loud, is there a more computational way to find these variations? Is there a negation filter (maybe exclusion) where if the two movies were overlaid you could negate one from the other, leaving behind the changes. What if the originals were modified so that one was just cyan and the other was just red, then when combined you would ignore the purples (or process them out) and focus on the left over cyan and red material. Problem is they probably don't align close enough, because of the widen frame composition...