OpenGame, Automatically Generating Games from Text

OpenGame, developed by researchers at the Chinese University of Hong Kong (CUHK) Machine Learning Lab (MMLab), is a system that generates a playable web-based game in about 30 seconds when a user enters the name of a desired game.

The system goes beyond simple code generation, operating as an agentic system. When a user enters text, the model writes HTML5-based source code, generates graphic assets, and defines the game rules. It then goes through a process of finding and fixing bugs on its own to produce a result that can be run immediately in a browser.

The current GitHub demo includes shooting, card battle, puzzle, and platform games using well-known IPs such as Star Wars, Harry Potter, and Marvel. Notably, it demonstrates the ability to accurately grasp the characteristics of a particular universe (e.g., Gryffindor, Mandalorian armor, etc.) from natural language descriptions alone—without any separate programming—and reflect them in the game elements.

However, the ability to generate games using copyrighted IP is expected to become an important issue going forward in relation to the application of copyright law to AI-generated content.