Portraits Editing using StyleGANs & Transfer Learning

The untangled latent space and the high-quality production style Generative Adversarial Networks has enabled a variety of image altering applications. For diverse tasks, new structures or task-specific training paradigms are frequently necessary. In this study, the spatial characteristics of Style GAN is investigated. It is proven that, without any additional architecture, tasks like image mixing, panorama creation, generation from a single picture, controlled and local multi - modal image to picture translation, and attribute transfer can be performed utilizing a pre-trained Style GAN that are equivalent to state-of-the-art approaches. There are currently a variety of models, primarily Generative Adversarial Network (GAN) for various image manipulation works. One such kind of GAN is Style GAN, because of which high quality image editing is possible. However not all tasks can be achieved using the style GAN method and to carry out those tasks some contiguous properties of style GAN have to be worked on. And with, a list of functions, such as blending of pictures, panorama creation as well as generation from a single image, controlled and local multimode image to image conversion, can be performed comparable to state-of-the-art algorithms. The proposed method is straightforward, effective, and efficient, and it may be used with any Style GAN model that has already been trained.