Local Video Analyzer

What It Solves

Video content is split across speech, on-screen text, scene changes, and platform signals, making manual transcription, clipping, highlight spotting, and editing planning expensive. video-analyzer-local turns local video files or downloadable links into structured, reviewable results, especially when you need offline analysis and clip-ready artifacts.

How It Works

After input, the tool processes audio and video in parallel: speech-to-text produces timestamped transcripts, and --diarize can separate speakers in conversations. The visual pipeline detects scene boundaries, recognizes objects, and runs OCR, while --slice-chapters outputs chapter clips with SRT and VTT files. It aligns sound, visual events, and text across the timeline to generate highlights, editing suggestions, EDL/CSV/JSON timelines, and HTML/JSON/Markdown reports. For Douyin, Kuaishou, Bilibili, and video-account links, it can extract metadata and analyze opening hooks, retention factors, pacing, and commerce-related keywords.

Boundaries

The tool requires ffmpeg and only accepts video files; pure audio, corrupted files, and executable files are rejected. Optional ASR models and OCR dependencies change accuracy and resource use; for long videos or low-memory machines, limit memory and disable visual/OCR passes. Speaker diarization, highlight detection, and scene cutting are still affected by audio noise, gradual transitions, and link availability, so outputs are best used as editing references rather than final decisions.