Brocacho Video Caption V3

Brocacho Video Captioner is a Dockerized AI agent built for Track 2 of the AMD Developer Hackathon: ACT II. It reads video-captioning tasks from /input/tasks.json, downloads each video, validates that its duration is between 30 seconds and two minutes, and extracts five evenly distributed midpoint frames using FFmpeg. These frames provide visual coverage from the beginning through the end of the clip while avoiding opening and closing fades. The agent uses MiniMax M3 through the Fireworks AI API to create a grounded description of the visible scene. Kimi then converts that description into four concise captions: formal, sarcastic, humorous-tech, and humorous non-tech. Each caption is restricted to one sentence of approximately 8 to 30 words, with reduced randomness to improve consistency and factual accuracy. The linux/amd64 Docker image includes the complete Python application and writes results to /output/results.json in the required evaluation schema. Failures are handled per task so one problematic video does not crash the complete batch.