The rapid advancement of machine translation (MT) has significantly improved text-based translation, yet challenges persist in specialized domains such as subtitle translation, which demands not only linguistic accuracy but also cultural sensitivity, technical compliance, and multimodal literacy. This study evaluates the performance of four automatic translation systems (ATSs)—ChatGPT-4o, DeepL, Google Translate, and Youdao Translate—by analyzing their subtitle translations for The Divine Damsel of Devastation, a promotional video for Genshin Impact. The research employs a dual evaluation approach: automatic assessment using nine MT evaluation metrics and manual error annotation and categorization based on linguistic, cultural, domain-specific, and technical dimensions. Results indicated that ChatGPT-4o consistently outperformed other systems, particularly in semantic accuracy and cultural equivalence, though all systems struggled with idiomatic expressions and literary references. The study not only reveals the current limitations of AI-driven and traditional MT systems in preserving cultural nuances but also proposes a comprehensive evaluation framework for automated subtitle translation, offering suggestions for improving ATSs in audiovisual contexts.