Video Captioning AI Agent

Video Captioning AI Agent is an end-to-end multimodal pipeline that automatically generates high-quality captions from short videos. Given a video URL and a list of requested caption styles, the system downloads the video, extracts representative frames using FFmpeg, performs scene understanding with a vision-language model, and generates captions tailored to the requested tone using a large language model. The pipeline separates visual understanding from language generation. The vision stage produces a structured description of the scene, including subjects, actions, setting, and key events. This intermediate representation is then passed to a caption generation model, allowing consistent caption quality while supporting multiple writing styles such as formal, sarcastic, humorous (tech), and humorous (non-tech). The project is fully containerized with Docker and follows the required input/output specification for the hackathon evaluation. It supports processing multiple captioning tasks, includes retry logic for external API calls, automatic cleanup of temporary files, and produces structured JSON output. By combining efficient frame extraction, multimodal reasoning, and style-aware language generation, the agent delivers captions that are both contextually accurate and tailored to the desired audience.