Changing video game graphic styles using neural algorithms

Byungho Yoo, Kyung-Joong Kim

Abstract: Recently, procedural content generation (PCG) has attracted positive attentions from gamers and applied for various content types such as maps, items and so on. Deep neural networks have been reported that they have potential to learn styles of artistic images. In this study, we propose to apply convolutional neural networks to change artistic styles of video game graphics. It's expected to change original games into different styles (modern, old-fashioned, scientific, and so on) given the input images. We applied the neural styling algorithm to the game images from Hedgewars, an open-source turn-based strategy game. Our results show that styles of video games can be changed from an input styling image.