📄 中文摘要
该研究提出了一种名为Multiverse的语言条件下的多游戏关卡生成器，旨在通过文本描述实现跨游戏关卡的融合。与以往仅限于单一游戏领域的文本到关卡生成器不同，Multiverse能够学习捕捉不同游戏之间结构关系的表示。该模型通过对齐文本指令和关卡结构的共享潜在空间，促进了多游戏之间的关卡生成。同时，基于阈值的多正例对比监督机制将语义相关的关卡连接在一起，从而增强了模型的生成能力和灵活性。该方法为程序化内容生成提供了更直观的控制方式，推动了多游戏环境下的关卡设计创新。
多元宇宙：通过共享表示实现语言条件下的多游戏关卡融合
📄 English Summary
Multiverse: Language-Conditioned Multi-Game Level Blending via Shared Representation
The study introduces Multiverse, a language-conditioned multi-game level generator that enables cross-game level blending through textual specifications. Unlike previous text-to-level generators that are limited to a single game domain, Multiverse learns representations that capture structural relationships across different games. The model facilitates level generation by aligning textual instructions and level structures in a shared latent space. Additionally, a threshold-based multi-positive contrastive supervision mechanism links semantically related levels across games, enhancing the model's generative capabilities and flexibility. This approach offers a more intuitive control over procedural content generation and advances innovation in level design across multiple gaming environments.