there is a probabilistic interactive component. this protocol allows the prover to fake it by just using a book where they know where Waldo is. in each iteration you choose whether to confirm it's the original book and page under the paper or see Waldo (but not both). a cheating prover has p = .5 of fooling a verifier.

but your concern is invalid to begin with. nothing in the definition of a zkp requires them to be multi-round interactive. there exist non-interactive zkp.