Need for empty ImageDatabunch for predict model?

It was mentioned in the lecture that we need an empty ImageDatabunch with no data at all for prediction. As in, create empty ImageDatabunch => create cnn with data => load from a previously saved model => do prediction. Why do we need an empty ImageDatabunch if we can are not doing a fit. We would have the same weight parameters if we don’t do a fit as well. Is it done just for saving the memory of the system?