Generating Gameplay-Relevant Art Assets with Transfer Learning

TL;DR

This paper introduces a CVAE-based transfer learning method to generate and modify game visuals that are relevant to gameplay mechanics, demonstrated on Pokémon sprites and types.

Contribution

It presents a novel CVAE system that incorporates transfer learning to generate gameplay-relevant art assets, addressing a gap in existing image generation methods.

Findings

Transfer learning improves visual quality and stability on unseen data

The approach effectively encodes gameplay mechanics into visual assets

Experimental results validate the method's potential for game development

Abstract

In game development, designing compelling visual assets that convey gameplay-relevant features requires time and experience. Recent image generation methods that create high-quality content could reduce development costs, but these approaches do not consider game mechanics. We propose a Convolutional Variational Autoencoder (CVAE) system to modify and generate new game visuals based on their gameplay relevance. We test this approach with Pok\'emon sprites and Pok\'emon type information, since types are one of the game's core mechanics and they directly impact the game's visuals. Our experimental results indicate that adopting a transfer learning approach can help to improve visual quality and stability over unseen data.