Ensemble video-caption agent tuned for style-match
A vision-model ENSEMBLE for video captioning: three frontier VLMs independently observe the frames, cross-verify, and one writer fuses them into four richly-detailed styled captions. 0.94 accuracy on the official clips.