Capcom Experimenting With Generative AI to Create 'Hundreds of Thousands of Unique Ideas' Needed to Build In-Game Environments

Capcom is exploring the use of generative AI to streamline the creation of the vast number of design concepts required for its game environments. This initiative comes amidst rising video game development costs, prompting publishers to explore AI tools – a move that remains controversial. Recent examples include a reportedly AI-generated cosmetic item in Call of Duty: Modern Warfare 3 and allegations of AI use in a loading screen for the same franchise. EA has even declared AI "the very core" of its business strategy.

In a recent interview with Google Cloud Japan, Capcom's Technical Director Kazuki Abe (known for his work on Monster Hunter: World and Exoprimal) detailed the company's AI experimentation. Abe highlighted the immense time and effort involved in generating the "hundreds of thousands" of unique design ideas needed for in-game assets. He cited the example of televisions, each requiring individual designs, logos, and shapes. This process involves numerous proposals, complete with illustrations and descriptions, for thousands or tens of thousands of objects per game.

To enhance efficiency, Abe developed a system leveraging generative AI. This system processes game design documents and generates design concepts, accelerating the development process. The AI also provides self-feedback, iteratively refining its output. Abe's prototype, utilizing models like Google Gemini Pro, Gemini Flash, and Imagen, has reportedly received positive internal feedback. The anticipated outcome is a substantial cost reduction and a potential improvement in overall design quality compared to manual creation.

Currently, Capcom's AI implementation is focused solely on this concept generation system. Other crucial aspects of game development, including gameplay design, programming, and character design, remain under the purview of human developers.