AnimeGamer: Infinite Anime Life Simulation via MLLM

This research paper introduces AnimeGamer, a novel method for creating infinite anime life simulation games. Unlike previous work limited to static images or finite gameplay, AnimeGamer uses a multimodal large language model to process user instructions and predict subsequent game states. These states include dynamic animation shots and updated character attributes, aiming for a more immersive and consistent gaming experience. The paper details the model's architecture, training process, and evaluation, demonstrating its superior performance in generating coherent and interactive anime game worlds compared to baseline methods.