For deep learning models, is seasonal differencing or just differencing a necessary step?

[ams015]

For many forecasting algorithms these are necessary preprocessing steps. But for deep learning I have not seen any clear guidelines or anything in the literature regarding this. I am playing around with the RNN based, MLP based and transformer based models.