CaptionHaki
CaptionHaki is an AI-powered Dockerized agent built for the AMD AI Hackathon Track 2. The system accepts a list of video captioning tasks, downloads each video, extracts representative frames, analyzes the visual content using multimodal AI models, converts the analysis into a structured JSON representation, and generates captions in multiple requested writing styles. The pipeline was designed to generalize across different categories of videos rather than memorizing the provided examples. It processes urban environments, animals, office scenes, nature, and other unseen content by extracting observable facts before generating captions. Separating visual understanding from caption generation improves consistency while allowing multiple caption styles to be created from the same structured description.
Category tags:
For Innovators & Creators