AI-Driven Dynamic Video Game Battles: Case Study of the Design of Tripo Clash

Real-time strategy card games are a rapidly growing genre that emphasizes strategic gameplay through real-time combat using collectible and upgradable cards. However, the reliance on a limited pool of pre-designed content can restrict long-term engagement. Additionally, procedural content generation for 3D asset generation also poses challenges due to its complexity and fidelity requirements. In response to these limitations, this paper investigates the incorporation of generative AI within the framework of Tripo Clash, a real-time strategy card game developed by the team. In this game, user choices are adaptively leveraged to influence gameplay visuals and dynamics, using Tripo for real-time 3D model generation. Simultaneously, large language models (LLMs) generate adaptive character behaviors, aligning content dynamically with player decisions. This synergy between Tripo and LLMs broadens the spectrum of available content and aligns closely with player preferences and gameplay styles, thereby enhancing immersion. A pipeline has been developed incorporating three critical components: (1)runtime 3D character generation, (2)dynamic character behavior logic creation via LLMs, and (3)customizable user operations in gameplay. The findings indicate that this approach effectively enriches game worlds and enhances player experience through dynamic, high-fidelity 3D content. Supplementary videos and code can be accessed at jw5ga183wsl.feishu.cn/docx/X8nfdh3yWoAtoKxfAqQcMRnGnBh