Nexus Vision
This project analyzes short video clips (30 seconds to 2 minutes) and generates captions in four required styles: formal (objective and factual), sarcastic (dry and ironic while staying accurate), humorous_tech (tech/programming humor grounded in real scene details), and humorous_non_tech (everyday humor with no jargon). The pipeline reads a task list from /input/tasks.json, processes each clip through a vision-capable Fireworks AI model to extract a factual scene description, then generates all four caption styles from that single grounded description — so every style describes the same real content, just in a different voice. Clips are processed in parallel (bounded by a concurrency limit), and results are written to /output/results.json in the required schema. The container is fully self-contained: it runs with no local files or manual setup beyond mounting input/output volumes and providing a Fireworks API key at runtime, completes a 13-clip local test set in under 2 minutes, and produces valid JSON on every run.
Category tags:.png&w=96&q=75){kind=small}
For Innovators & Creators