GameGenesis: A Multimodal AI Revolution in Board Game Design
Disciplines
Artificial Intelligence and Robotics
Abstract (300 words maximum)
This project aims to build an AI model that automates the creation of novel and unique board game mechanics as well as the board and game pieces themselves. This system leverages established NLP’s and computer vision models to translate a user-provided board game description into a complete game design package. Specifically, it will combine text-to-rule generation with text-to-image synthesis to produce coherent and visually relevant game components. For the text-based generation of game mechanics, the system utilizes GPT 4 on a carefully constructed dataset that includes rules, reviews, and possibly even designs, which are sourced from platforms like BoardGameGeek, subsequently training the model to output rule sets, including objectives, resource management, turn-based mechanics, and the winning conditions. These game mechanics are engaging, logical, compelling, and simply fun. In conjunction, this AI system incorporates a visual design element that can generate aesthetically and logically relevant game boards and pieces. Contemporary models like DALL-E can be fine tuned on a dataset of established board game imagery, which allows the model to generate design elements consistent with the board games we all know and enjoy. Because this model uses multiple modalities, CLIP (Contrastive Language–Image Pre-training) is used to ensure that the designs match the user text prompt and prospective game rules. The anticipated outcome is the creation of a comprehensive model that can reduce the time, effort, and expertise that is required to create engaging, balanced, and fun board games. By leaning on AI, this project seeks to create a system that can be utilized and applicable in both independent and commercial game development.
Academic department under which the project should be listed
CCSE - Computer Science
Primary Investigator (PI) Name
Md Abdullah Al Hafiz Khan
GameGenesis: A Multimodal AI Revolution in Board Game Design
This project aims to build an AI model that automates the creation of novel and unique board game mechanics as well as the board and game pieces themselves. This system leverages established NLP’s and computer vision models to translate a user-provided board game description into a complete game design package. Specifically, it will combine text-to-rule generation with text-to-image synthesis to produce coherent and visually relevant game components. For the text-based generation of game mechanics, the system utilizes GPT 4 on a carefully constructed dataset that includes rules, reviews, and possibly even designs, which are sourced from platforms like BoardGameGeek, subsequently training the model to output rule sets, including objectives, resource management, turn-based mechanics, and the winning conditions. These game mechanics are engaging, logical, compelling, and simply fun. In conjunction, this AI system incorporates a visual design element that can generate aesthetically and logically relevant game boards and pieces. Contemporary models like DALL-E can be fine tuned on a dataset of established board game imagery, which allows the model to generate design elements consistent with the board games we all know and enjoy. Because this model uses multiple modalities, CLIP (Contrastive Language–Image Pre-training) is used to ensure that the designs match the user text prompt and prospective game rules. The anticipated outcome is the creation of a comprehensive model that can reduce the time, effort, and expertise that is required to create engaging, balanced, and fun board games. By leaning on AI, this project seeks to create a system that can be utilized and applicable in both independent and commercial game development.