In this short note, recent results on the predictive power of kernel principal component in a regression setting are extended two ways: (1) model-free setting, we relax conditional independence model assumption to obtain stronger result; and (2) is also infinite-dimensional setting.