A Convolutional Model to Generate Human and Anime Faces Using DCGAN

Generative Adversarial Networks are one of the primary sources of booming Artificial Intelligence-based Art generation. Nowadays Generative Adversarial Networks possess many applications including Text to Image Translation, Face Frontal View Generation, and Generating fresh human poses. GANs have been extensively used for image synthesis in recent years, including generating synthetic human and anime-style images. In this paper, a modified GAN architecture has been proposed to generate human faces and anime images that are visually appealing, detailed, and diverse. This paper describes the architecture of a GAN model, which incorporates a combination of a generator model and a discriminator model, and the training procedure using Anime Face Dataset and CelebFaces Attributes Dataset available. The evaluation of the execution of the suggested GAN model is performed using various metrics, including visual inspection, loss functions and the real and fake score of the discriminator model. The outcomes demonstrate that proposed working GAN design can generate good-quality anime and human face images that are visually similar to real images and exhibit diversity in terms of style, character, and background.