Media Specificity of Generative AI Video from the Perspective of Remediation

Generative AI technology has expanded from text and image generation to the domain of video, with recent advancements such as OpenAI’s Sora and Google DeepMind’s Veo2 delivering highly sophisticated results that surpass existing video technologies. These technologies evoke a strong sense of realism, enabling viewers to experience reality directly without perceiving the presence of the medium. Despite these advances, however, the artificiality or imperfections evident in the generated outputs simultaneously indeed prompt an awareness of the medium itself. This dual aspect arises from large-scale dataset training, complex deep learning architectures, and algorithm-driven automated generation processes, revealing the distinctive mediality of generative AI video. This phenomenon is closely connected to Bolter and Grusin’s concept of remediation, which describes how new media types are justified by transforming and reconstructing the forms or logic of existing media, employing the simultaneous strategies of immediacy and hypermediacy. Based on this remediation theory, this study regards generative AI video not merely as a content creation tool but as an independent media form, analyzing its medial characteristics. The analysis indicates that generative AI video simultaneously employs both immediacy and hypermediacy strategies and performs remediation focused more on ‘content’ than on ‘form.’ Furthermore, through a self-referential structure where AI-generated videos are reused as training data, generative AI video evolves into a recursive media form that iteratively renews itself. This discussion offers a theoretical foundation on which to grasp the medial nature of generative AI video and aims to contribute to future discourse and practical applications in media theory, digital content design, and AI-driven creative practices.