We generate synthetic images with a generative adversarial network (GAN) model trained with image patches from specular microscopy corneal endothelial cells. Preliminary results show it may be a suitable approach for reliable cell segmentation.