An Art Font Generation Technique using Pix2Pix-based Networks

Minghao Xue, Yasuaki Ito, Koji Nakano


Art font is a typeface decorated with special visual effects and is widely used in two-dimensional graphic design. Generating art fonts directly from images is a convenient way to design art fonts that reduce the need for expertise. The appearance of such decorative fonts is similar to the foreground of the reference image. In this paper, we propose an art font generation method using machine learning, which uses multiple Pix2Pix networks to quickly generate high-quality art fonts similar to the foreground of a reference image. In this method, first of all, three Pix2Pix networks are used directly. One of the networks is used for mask dataset generation and the other two networks are used for art font generation. Secondly, the parameter β can be used directly to control the effect of the network generation without changing the network structure of Pix2Pix. The experimental results show that the proposed method effectively controls shape change and texture imposition by dividing them and can generate artistic art fonts. Furthermore, we showed that the proposed method could generate artistic and complicated art fonts for input fonts with different shapes, indicating that the method has high practical utility.


art font; Pix2Pix; Generative Adversarial Network; machine learning

Full Text:



  • There are currently no refbacks.