The article explores a method to automatically transform Korean fonts into different styles using a combination of Unbalanced U-net and Generative Adversarial Networks (GAN). The researchers developed a new architecture called Unbalanced U-net, which maintains the meaning and structure of the original font while changing its style. By using a complex loss function in GAN, the model showed faster convergence and stable training loss compared to the traditional Balanced U-net. This approach successfully addressed the performance degradation issue in font transformation tasks.