FrameTone
FrameTone is a competition-ready video captioning agent built for Track 2 of the AMD Developer Hackathon: ACT II. It turns any 30-second to two-minute clip into four English captions: formal, sarcastic, humorous-tech, and humorous-non-tech. Each version keeps the same factual core while changing voice clearly, so humor never comes at the cost of accuracy. FrameTone is designed for varied unseen content, including people, animals, nature, cities, food, sports, weather, and technology. It uses Fireworks AI inference on AMD-backed infrastructure and ships as a public linux/amd64 container that follows the exact input and output contract. In validation with the three official 4K practice clips, the container returned accurate, clearly differentiated captions for every required style. The result is a reliable, expressive captioning experience that is ready for the hidden judging set.
Category tags:
For Innovators & Creators