Slide between two faces, A and B. Mixing their codes and decoding the mix gives new in-between faces that were never in the training data.
Encode two inputs, mix their codes, decode the mixture. Every point along the way is a new output that was never in the training data.
