Published Paper Details:

One of the critical multimedia challenges in today's digital landscape is the rapid and effective creation of video content enriched with relevant metadata. This metadata is essential for improving discoverability, organization, and engagement, but the process of manually generating it is both time-consuming and inconsistent. To address this, we propose a Video Metadata Automation System that utilizes recent advancements in artificial intelligence--particularly large language models (LLMs)--to automatically generate key video metadata, including titles, descriptions, captions, and thumbnails. Our system employs an end-to-end automated framework that integrates techniques such as audio extraction, speech-to-text transcription, and video analysis. Using APIs for connection to high-performing LLMs, the extracted transcripts are used to create metadata in a structured and SEO-friendly format. The most suitable LLMs are evaluated and selected for optimal performance in this context. Furthermore, the system features integration with image generation models such as Stability AI's Stable Diffusion to generate thumbnails based on the video's context. This paper explores the effective utilization of open-source tools and state-of-the-art AI technologies, presenting a unified and modular solution that simplifies the video content creation pipeline. By automating the metadata generation process, our system significantly reduces the workload for creators, enhances video accessibility, and enables faster content publication with minimal manual intervention.