SceneScribe

SceneScribe is an AI video captioning agent built for the AMD Developer Hackathon Act II. It transforms any video into four distinct caption styles, empowering creators, marketers, and developers to instantly generate captions that match their audience, platform, and brand voice. The project uses a two-stage AI pipeline. First, a vision model analyzes key frames to generate a neutral, factual scene description. Then, a language model rewrites that description into four distinct tones: Formal, Sarcastic, Tech-Humorous, and Everyday-Humorous. What makes SceneScribe unique is its intelligent frame selection. Instead of taking fixed interval frames, it uses histogram analysis to detect scene changes and motion detection to identify action peaks, ensuring the AI sees the most meaningful moments in the video. The project also features confidence scoring, which allows the AI to express how certain it is about each caption. A self-verification engine checks factual consistency between the captions and the original scene, regenerating only the captions that need improvement rather than restarting the entire pipeline. This ensures high-quality, factually accurate outputs in every style. The application is packaged as a Docker container for easy deployment and evaluation, with a polished frontend built in React that provides a premium user experience with glassmorphism design, responsive layouts, and real-time processing status updates. Built as a solo submission in five days, SceneScribe demonstrates the power of combining intelligent video processing, state-of-the-art AI models, and thoughtful user experience design.