Researchers at the Chinese University of Hong Kong's MMLab have developed an open-source AI called OpenGame, which can generate entire playable video games from text prompts. The process involves identifying the game type, creating a project framework and generating necessary assets.
The demos include an Avengers-style platformer, a Harry Potter card battle game, and a survival game inspired by Netflix's Squid Game. This approach contrasts with other AI models that generate individual code snippets or assets.
OpenGame uses a specialized model called GameCoder-27B, trained on game development workflows through reinforcement learning. It can produce over 150 different types of games and includes an evaluation system to ensure the final product meets initial prompts before deeming it playable.
However, some developers remain skeptical about its real-world utility, pointing out the limitations in persistent state or multiplayer functionality and concerns about resource costs. The AI framework’s ability set evolves over time through a history of patterns, designs, and fixes.
