No, we just need to train a neural network to generate cat pictures from a small set of inputs (like a uint8 for pose, a uint8 for breed, etc.) and then just transmit those parameters instead of the entire picture.