Video Analyzer

Extracts frames at adaptive rates, creates montage grids, detects scene changes, and dispatches parallel subagents for visual and audio analysis.

Gotchas

• Social media downloads (Instagram, TikTok) embed a MJPEG thumbnail as video stream 0. The scripts auto-detect the real H.264/H.265 stream and skip attached_pic streams.
• macOS grep lacks -P (Perl regex). All scripts use sed and basic grep for compatibility.
• FFmpeg treats . in output filenames as image sequence patterns. Key frame extraction uses -update 1 to write single images.
• Portrait videos (height > width) use 3x5 grid layout instead of 4x4 for better cell visibility.
• Scene detection threshold is 0.3. For static content (presentations, documents), lower to 0.1. For fast action (games, sports), raise to 0.4.
• Always verify extraction produced output before dispatching analysis agents.

Prerequisites

Requires ffmpeg, ffprobe, python3, bc. For transcription: whisper CLI.