So you just need a GAN trained with a dataset that excludes earrings and glasses. Then maybe another GAN trained to add earrings and glasses onto faces that don't have any :)