The problem is one of discrete values. You are asking about "a boy", but there are no "a boy"s. There are only fractions of boys. The total sum of "boy fractions" is the same, but the number of pairs of "boy fractions" that meet the distinction of "a boy" changes. "A boy" does not vanish because there were never "a boy"s in the first place.
One might have better luck with "boy heads", and you can see that at roughly 5 oclock there is a "boy head" that turns into a "boy arm". So part of the trick is that some of the "boy fractions" change in your interpretation. Without the trick, you would reasonably say, "Wait, there is still a boy head there! So there are still 13 boys!"
If done right, the idea is that in the one configuration you have thirteen 24/25ths of a boy, so thirteen 96%-boys. You hide this as each one being slightly skinny.
In the other configuration you have twelve 26/25ths of a boy, so twelve 104%-boys. You hide this as each one being slightly fat.
The danger is that if you make the drawing too detailed and gorgeous, someone will be able to look at a little detail like eyes or so, if one of these 4% slices contains an eye then you may end up with boys with 1 or 3 eyes, something that any looker would say spoils the illusion.
You have a couple options there.
- People will accept a "dead zone" where the two moving parts interact, you can try to locate an eye inside the "dead zone" so that you don't trigger this W-T-F moment.
- Use cartoonishness/ambiguity. So maybe this dot means "eye" on this cartoon but "freckle" on that cartoon, similarly by locating the exact eye on the border between the two, maybe half a line goes from being "long eyelashes" to being "the middle of a winking eye" or so.
In this view, the answer to your question is 'one of each pair, except in one case in position A only.'