Film Production Through a Korean-Specific Text-to-Video AI Model

This study outlines the development of a Korean-specialized text-to-video generative artificial intelligence model. In the Korean film industry, the traditional pipeline of pre-production, production, and post- production varies in efficiency in terms of capital investment and labor input, and is thus greatly influenced by capital. Within this structure, the gap between commercial films with large capital investments and lower-budget productions – such as independent films and art films – has widened, ultimately compromising cinematic diversity. In addition, owing to the limited availability of video generation technology that accepts Korean input, video content creators in Korea utilize existing video generation technology inefficiently. To address these issues, we developed a Korean-specialized text-to-video generative AI model, and proposed efficient utilization methods for the Korean film industry to break free from its existing capital-centric structure. We produced short films with runtimes within three minutes, confirming the ability to create video content for pre-visualization in the pre-production stage, as well as additional insert cuts required in later stages. Nonetheless, further research is required to enhance this technology and achieve more polished scenes by maintaining consistent object shapes and reflecting various shot sizes and camera movements.