Scene-change thresholds: review markers before calling them shots

Define the detection purpose and source interval

Retain the authorized source and identify the video stream and interval to inspect. Decide whether the task is editorial shot indexing, representative thumbnail selection or a processing diagnostic. The acceptance criteria differ: a useful reference image does not automatically establish the precise boundary between two shots.

Choose a passage containing known hard cuts, motion and lighting changes where possible. Record the actual source version. A retimed or filtered derivative can produce different scores from the parent, so a marker list without its source identity is not reproducible evidence. Do not use one quiet passage to claim that an entire recording has no shot changes.

Select a bounded native diagnostic and keep its timestamps

A scoped example is ffmpeg -t 30 -i source.mp4 -map 0:v:0 -vf "select='gt(scene,0.4)', showinfo" -an -f null -. The input duration is placed before -i to bound the read even when no frames meet the threshold. It selects frames whose scene score exceeds the illustrative threshold and reports selected-frame information without writing a finished movie. The number is a test setting, not a universal recommended boundary detector.

Retain the relevant reported positions with the source and command. Inspect the filter result and actual processing interval before using them. Changing the threshold changes the candidate set rather than making every marker true. Do not distribute raw diagnostic logs containing unnecessary private paths or unrelated source details when a scoped marker record is sufficient.

Verify candidates and missing transitions against the movie

Watch before and after each selected position and classify the event: hard cut, flash, exposure change, moving object or another visual difference. Review at least some known transitions that were not selected so the task does not measure only false positives. A changing threshold can trade missed boundaries against extra markers without resolving every transition type.

If the accepted markers will drive cuts, thumbnails or a storyboard, inspect that actual derivative’s boundaries and content too. A selected frame may be useful without being an exact edit point, and extracted images do not retain the soundtrack or full context. Preserve the source and record the verified purpose and uncertain markers. SocialVidDownloader’s storyboard and thumbnail utilities produce their actual documented outputs; they do not certify this scene-threshold analysis as a complete editorial shot map or authenticity record. Keep automatic candidates distinct from reviewed conclusions.

Worked example

Example: a camera flash exceeds the selected score while a slow dissolve does not. The reviewer records the flash as a lighting event and checks the dissolve separately. The marker list remains a diagnostic aid instead of being published as a fully verified list of scene cuts.

Related guidance

• A selected thumbnail versus a representative contact sheet
• Detected black or frozen intervals need recording context

Sources and verification

• FFmpeg select scene scores and showinfo

Reviewed 2026-10-13. Product behavior is limited to available source formats and provider access; it is not a promise that every public link works.