Style-Aware Video Captioning Agent
Style-Aware Video Captioning Agent is an end-to-end system for Track 2 video caption generation. The agent reads video tasks, downloads each clip, samples representative frames, removes near-duplicate frames, and converts the visual sequence into compact frame grids for multimodal analysis. A vision-language model first produces a factual chronological description of the video, preserving visible subjects, actions, setting changes, and camera behavior. That structured analysis is then passed into a second language-model stage that generates captions in each requested style: formal, sarcastic, humorous with technology references, and humorous without technical jargon. The project is designed to separate visual understanding from style generation. This keeps the captions grounded in the actual video while allowing flexible tone control. The final output follows the required Track 2 JSON schema, producing one captions object per task with all requested styles.
Category tags:
.png&w=96&q=75){kind=small}
.png&w=96&q=75){kind=small}
.png&w=96&q=75){kind=small}
For Innovators & Creators