Understand an mp4 video by extracting in-focus frames and reading them with the running model. Downloads a URL or reads a local file, samples frames with ffmpeg (e.g. 1/sec), scores sharpness with variance-of-Laplacian blur detection, picks the sharpest frame per time window (checking neighbors when one is blurry), then reads those frames to answer the request. Use when the user asks to review, analyze, summarize, describe, or answer questions about an mp4 / video / screen recording / clip, or asks "what happens in this video".
