MC-GAN: A Reimplementation of GameGAN With a Gaming Perspective

Abstract

Creating games is a costly process involving tons of vigorous efforts, time, and resources. Different game engines were introduced to facilitate the game-making process, such as Unreal Engine, Unity, and Godot. However, despite the advent of game engines, the vacuum of automatically creating new games by computers was still felt. With artificial intelligence (AI) development and its growing presence in the industry, game developers made a new game development branch using AI. One of the essential steps in this context was the introduction of GameGAN, which inspired us for this article. Here we propose Multi-Class GamgeGan(MC-GAN), a starting point for the next generation of game engines based on GameGAN. In addition, MC-GAN is capable of classifying each desired game element and even changing its class to change the element’s nature (e.g., MC-GAN can change static elements to dynamic ones). Moreover, MC-GAN uses only one complex model trained once per game. Once it is trained for that specific game, there will be no need for training to produce a new level of that game, a considerable step in the developing industry. Considering that MC-GAN is only the beginning of this big journey; therefore, it is now the best choice for developing web games since they are compact and straightforward. This article first briefly introduces GamGAN and its features; afterward, we get through MC-GAN features and compare them to other similar works.