Multi-Style AI Video Caption Generator

Our project is an AI-powered video captioning system that transforms videos into engaging captions tailored to different audiences and use cases. Given a video URL, the system automatically downloads the video, extracts representative frames, analyzes the visual content using a multimodal large language model, and generates captions in multiple styles from a single inference pipeline. The generated captions include formal, sarcastic, humorous tech, and humorous non-tech styles, allowing creators, marketers, educators, and social media users to quickly produce content suitable for different platforms and audiences without manually rewriting captions. The application is designed as an automated pipeline. It processes a JSON input containing one or more video tasks, extracts key frames from each video using FFmpeg, summarizes the scene with an AI model, and produces structured JSON output containing captions for every requested style. This enables batch processing of multiple videos efficiently. The project emphasizes speed, scalability, and ease of deployment. It is fully containerized using Docker, making it portable across development and production environments with minimal setup. The modular architecture separates video processing, frame extraction, AI inference, and output generation, making the system easy to maintain and extend. Potential applications include social media content creation, marketing campaigns, accessibility, video archiving, educational content, and automated media workflows. By generating multiple caption styles from the same video, the system saves creators significant time while improving audience engagement through style-specific messaging.