Does GMM with first-order differencing require stationarity of endogenous variables?

I'm trying to estimate a panel VAR model with first-order differenced GMM estimator in R, using the {panelvar} package.

I know that GMM requires the dependent variable to be stationary. But Does it still matter in a GMM estimation with first-order differencing transformation?

Of course, first-order differencing doesn't necessarily assure stationarity. But when first-order differencing can remove unit root from the dependent variable, can I give the original dataset (without differencing beforehand) to the GMM estimation?