In calculating the amplitude for the process $e^-\gamma\to{}e^-\gamma$ the substitution $\sum\epsilon_{\mu}\epsilon^*_{\nu}\to-\eta_{\mu\nu}$ is useful to sum over photon polarizations.
If we instead consider the process $e^-e^+\to{}2\gamma$ our amplitude will be of the form $M=\epsilon^*_{\mu}\epsilon^*_{\nu}M^{\mu\nu}$. Is the prescription $\sum\epsilon^*_{\mu}\epsilon^*_{\nu}\to-\eta_{\mu\nu}$ still valid? In case it is not, is there a similar expression to carry out the photon polarization sum and how would this formula be justified?