>>47074
When you give a target image to these generators you do not 'train' an AI as such. They turn the target image into parameters that could be used in facial recognition - width of chin, length of nose etc - then use the error function of the recognition AI to determine how to change a second input image until it gets recognized as the target.

None of the AI parameters are changed. To actually train an AI would typically take 100,000+ images.