Submission Title: SceneScribe AI
SceneScribe AI is an AI-powered video captioning agent that analyzes video content across multiple temporal frames and generates captions in four requested styles: formal, sarcastic, humorous tech, and humorous non-tech. The system extracts uniformly distributed frames with FFmpeg, builds a grounded factual representation of visible objects, actions, scene context, and temporal changes, and then transforms that representation into style-specific captions. The agent is packaged as a Linux AMD64 Docker container, reads tasks from /input/tasks.json, processes each video URL, validates all requested output styles, writes valid JSON to /output/results.json, and exits automatically. The pipeline has been tested successfully on the three official example videos covering urban traffic, an animal scene, and a person working in an office. It uses MiniMax M3 through Fireworks AI for multimodal video understanding and caption generation.
Category tags:
.png&w=96&q=75){kind=small}
For Innovators & Creators