버추얼 캐릭터 실시간 운용 방식의 유형과 특성 연구

초록

(Background and Purpose) As real-time rendering technologies and generative artificial intelligence (AI) continue to advance, the virtual character industry is growing rapidly and diversifying across various content domains, including entertainment, live commerce, and AI streaming. However, existing studies have largely focused on specific technical implementations or fragmented case analyses, with limited research systematically classifying virtual characters using real-time operation methods. Accordingly, this study aims to classify virtual characters based on real-time operation methods and analyze, from a spatial design perspective, the operational characteristics of each type and the differences in the digital spatial experiences they create. (Method) This study adopted a case study approach. The analysis was limited to domestic and international virtual character cases released between 2022 and 2026 whose production processes could be verified through publicly available sources, including official articles, academic papers, producer interviews, and official websites. Cases were classified into three types—the Real-time Motion Operation Type, the Pre-produced Operation Type, and the AI Autonomous Operation Type—and analyzed based on production year and country, content type, content details, real-time operation technology, operating agent and level of human involvement, and operational characteristics. PLAVE was selected as the representative case for the Real-time Motion Operation Type, Zanmang Loopy Virtual Character Live for the Pre-produced Operation Type, and Neuro-sama for the AI Autonomous Operation Type. (Results) The Real-time Motion Operation Type demonstrates high immersion and spontaneous interactivity through direct real-time mapping of human operator data, but is constrained by high costs and labor requirements owing to the need for specialized equipment and skilled personnel. The Pre-produced Operation Type secures consistent character expression and stable output quality through controller-based execution of pre-produced action data but is limited in its capacity for spontaneous responses beyond the pre-produced action range. The AI Autonomous Operation Type enables autonomous interaction and long-duration broadcasting without operator intervention through large language model (LLM)-based real-time inference but requires ongoing monitoring to maintain output consistency and manage unpredictable responses. (Conclusions) The three types are independent operation methods that demonstrate their respective strengths based on content purpose and operational context rather than representing a hierarchy of technological advancement. The Real-time Motion Operation Type is suited for fandom-based entertainment content, the Pre-produced Operation Type for content requiring repetitive and stable operation such as live commerce and brand collaborations, and the AI Autonomous Operation Type for emerging content formats such as AI streamers. This study systematically classified virtual characters based on real-time operation methods, analyzed the technical and industrial characteristics of each type, and examined the differences in the digital spatial experiences created by each operation method from a spatial design perspective, thereby providing a foundation for the strategic application of virtual character technologies in the content production industry.

키워드