Over the past decade, the skyrocketing cost of producing AAA video games has become one of the gaming industry’s biggest challenges. In 2000, a dozen specialists could make a big hit for the PlayStation. Today, however, creating a new version of any franchise for a game console or PC requires several years of work by hundreds of artists, designers, and programmers. Even the processes of creating casual games like item hunters and puzzle games take many months of work, from conceptualization to release on the Apple App Store or Google Play.
One of the biggest costs in game production is the creation of the game asset. Players want the content of games to be not only interesting, but also unique. Characters, textures, skins, exotic locations, and various in-game boosters all have to be carefully customized, refined, and worked on by artists, and this is incredibly expensive.
Another challenge is scaling. In order to stay competitive, game studios need to release a lot of DLC updates and assets. This is especially true today when all monetization trends are shifting to in-app purchases such as character skins, accessories, vehicles, emotes, and maps. Talent is not enough. One solution is to use outsourced game art studios with low wages, but finding a reliable studio and building a long-term partnership is not an easy task.
That’s why studios are always looking for technologies that can help reduce the cost of game development. Recent advances in neural image generation models such as DALL-E, MidJourney, and StableDiffusion give hope that the realization of this dream may not be that far away.
Introducing Nvidia’s GET3D, the latest discovery in this field that was recently announced in an article. This AI model was trained using only 2D images. The model is capable of generating 3D shapes with high quality textures and complex geometric details. The objects that can be created are quite diverse: vehicles, characters, animals, people, buildings, various open spaces that can be combined into entire cities with their inhabitants. The export formats are suitable for most popular graphics software, making it easy to import the shapes into 3D visualization tools and game engines, and they can be easily incorporated into existing art production workflows.
How close are we to actually implementing this and similar models in real-world game development? “The technology presented is impressive, and we will surely see a lot of generated art objects in games soon. However, they will not replace the creative team of artists in the foreseeable future, as AI still requires well-designed input data to produce quality results and careful post-processing to make the processes consistent and flawless,” expressed their expert opinion of the specialists of the game art studio Absolutist. It seems that in game development, as in many other fields, machine learning applications are very useful as an aid to artists and designers, but not a substitute for human creativity and skills.