The invention provides a facial image conversion method based on a cycle generative adversarial network. The main content of the method comprises a Wasserstein generative adversarial network (WGAN), astructural similarity (SSIM) loss, background subtraction method and face mask and a generative adversarial network (GAN). The method comprises the steps of using a generator network and a discriminator network to compete against each other, forming the cycle GAN by using a traditional GAN loss function and a new cyclic consistency loss function, improving the WGAN, improving the training of the GAN through loss, matching the SSIM loss and brightness, contrast and structural information of a generated image and an input image, inputting a binary mask and the images during training, and applying an element product to reconstruct the loss. According to the method, the cycle generative adversarial network is used, the method has higher consistency and stability in converting facial expressions, facial details and edge details can be processed well, and thus a converted image is more natural and more realistic.