A two-stage video captioning agent that samples keyframes, generates a faithful scene description with a vision-language model, then rewrites it into four distinct tones — formal, sarcastic, humorous-tech, and humorous-non-tech — via the Fireworks AI API.