Portrait illustrations are illustrations or paintings depicting full-body or bust-up portraits of people. They have been used for a wide range of purposes, including fan art of characters in anime and manga, illustrations and book covers for novels, social games, and advertisements. In portrait illustrations, various backgrounds are drawn behind the main characters. Among them, there can be found a characteristic style of backgrounds in which geometric figures such as circles, rectangles, and triangles are arranged on a flat surface. This specific style gives the impression of relative simplicity and sophistication, while at the same time enhancing the main subject of the human figures. In addition, depending on the layout and color scheme of the figures, a variety of background patterns can be created, allowing for a wide range of expression. On the other hand, adjusting the layout and colors to enhance the main subjects requires a great deal of effort, experience, and artistic sense.

In this thesis, these backgrounds are named complementary backgrounds, and an interactive genetic algorithm is proposed to generate them semiautomatically. The design of complementary backgrounds can be attributed to an optimization problem in terms of colors, texture, shapes, and layout of figures. In the interactive genetic algorithm, the user is allowed to evaluate the fitness of the background through repeated gene manipulation, which enables the optimization process to reflect their intention. In addition, by mutating part of the genes or crossing genes, it is possible to generate a wide variety of background patterns without converging to a local optimum. In the thesis, a system that assists users in interactively generating complementary backgrounds was developed. In the proposed system, the user mainly evaluates the fitness, and views the characteristics of the population and changes caused by the genetic variations and repeats the genetic manipulation according to the situation.

This research enabled the creation of complementary backgrounds that are effective in enhancing a given character illustration. Evaluation experiments showed that the system can generate backgrounds that enhance characters more than a generative AI. The system was also found to be able to generate complementary backgrounds that reflect the user’s own preferences while stimulating the user’s senses.


NameAffiliationWeb site
Tomoya KobayashiKeio University


  1. Grant-in-Aid for Challenging Research (Pioneering): 20K20481 (2020―2023)

Back to CA team page