Text2AC: A Framework for Game-Ready 2D Agent Character(AC) Generation from Natural Language

The visual appearance and personality of game characters are crucial to the player experience. Assembling characters in 2D games is a common practice with a significant demand for creativity, yet the creation process remains challenging, confined to professional graphic designers and animators. Traditional automated character creation in games usually involves manual parameter adjustment or the use of photos as references, coupled with the selection of pre-set backgrounds or professions within the game, leading to limitations in terms of freedom, usability, and consistency. However, with advancements in natural language processing, large language models (LLMs), and generative AI, we have the opportunity to reimagine the process of user-interactive character creation. This paper introduces a novel text-to-character methodology, integrating LLM technology with generative image AI, eliminating the need for reference photos or manual parameter editing while providing extensive options for character creation. Through natural language processing and automated skeletal animation, the generated 2D characters are capable of being driven across various web platforms and game engines. This exploration aims to establish a framework for driven agents that harmonizes the visual layer with linguistic personality, thereby providing a more enriched user experience in games and other applications as agents.