Yes, you have to be consistent and perform the same preprocessing for your train and test dataset.
If you use the Deployment API instead of a downloaded .h5, the same data processes that have been done during training will be performed automatically. That is, you don’t need to do anything.