Recently, generative adversarial networks (GANs) have demonstrated high-quality reconstruction in face completion. There is still much room for improvement over the conventional GAN models that do not explicitly address the texture details problem. In this paper, we propose a Laplacian-pyramid-based