Recomposing the Pokémon color palette

Abstract

In digital games, the visual representation of game assets such as avatars or game levels can hint at their purpose, in-game use and strengths. In the Pokémon games, this is particularly prevalent with the namesake creatures’ type and the colors in their sprites. To win these games, players choose Pokémon of the right type to counter their opponents’ strengths; this makes the visual identification of type important. In this paper, computational intelligence methods are used to learn a mapping between a Pokémon’s type and its in-game sprite, colors and shape. This mapping can be useful for a designer attempting to create new Pokémon of certain types. In this paper, instead, evolutionary algorithms are used to create new Pokémon sprites by using existing color information but recombining it into a new palette. Results show that evolution can be applied to Pokémon sprites on a local or global scale, to exert different degrees of designer control and to achieve different goals.