I can't go into depth on our models or data sets, because this is a cat and mouse game between creation/detection. We create deepfakes using all known tools, currently non-public methods as well as our own methods.