What this tool measures
Load one local video to scan sampled frames and audio for scene changes, black or near-black intervals, duplicate or near-duplicate intervals, and silence. The analyzer only reports findings: it does not write, re-encode, trim, or remove anything from the source. The selected video is not uploaded or sent to a server; processing runs in your browser.
What the analysis settings control
These controls automatically refresh the current report. The values below are the available ranges and defaults.
- Sample interval (ms) - controls how often frames are sampled. The range is
50-5000; the default is500. A lower value covers the timeline more closely but requires more work. - Scene change sensitivity (0-100) - sets the perceptual-hash distance threshold for scene changes. The range is
1-100; the default is35. - Black frame luminance (0-255) - sets the average sampled luminance at or below which a frame counts as black or near-black. The range is
0-255; the default is16. - Black frame min duration (ms) - filters out shorter black or near-black runs. The range is
0-60000; the default is200. - Near-duplicate sensitivity (0-100) - sets the maximum perceptual-hash distance for near-duplicate frames. The range is
0-100; the default is5. - Duplicate min duration (ms) - filters out shorter duplicate or near-duplicate runs. The range is
0-60000; the default is500. - Silence threshold (dBFS) - sets the audio level at or below which samples count as silent or near-silent. The range is
-90-0; the default is-50. - Silence min duration (ms) - filters out shorter silent or near-silent runs. The range is
0-60000; the default is300.
How each result is detected
Scene changes compare consecutive sampled frames with a 64-bit grayscale difference hash and Hamming distance. A result includes confidence derived from the normalized distance.
Black intervals use the average sampled luminance. Consecutive samples must meet Black frame luminance (0-255) and Black frame min duration (ms). Their reported boundaries are estimates based on the sampled frames and interval.
Exact duplicate intervals use a checksum from downsampled grayscale samples with quantized luminance. This is an exact match of the tool's generated checksum, not proof that the raw video frames are byte-for-byte identical. Near-duplicate intervals use perceptual-hash similarity and can include visually similar frames that are not pixel-identical.
Silence analysis measures short RMS audio windows as dBFS. Stretches at or below Silence threshold (dBFS) that meet Silence min duration (ms) are reported with their peak level. Black, duplicate, and scene events include frame thumbnails; silence events provide timestamps and metrics without thumbnails.
How to analyze a video
- Drop a video file into the drop zone, or click browse files.
- Adjust the eight settings if the defaults do not fit the video.
- Analysis starts automatically after the video metadata is ready and whenever a setting changes.
- Review the four event panels. Click an event row to seek the player to its start time.
- Choose Export JSON report for the full report, or Export CSV report for event rows and fields such as type, timestamps, duration, metric, confidence, and hash type.
Limitations and failure cases
Detection evaluates sampled frames, so a cut or black flash shorter than the sample interval can fall between samples. Analysis also depends on the browser decoding the video and audio. If video analysis fails, the tool shows an Analysis failed alert. If the file has no audio track or the browser cannot decode it, the silence panel shows No decodable audio stream was found in this file.
Frequently asked questions
Does this tool upload my video anywhere?
No. The selected video is not uploaded or sent to a server; frame and audio analysis run locally in the browser.
Does it modify my file?
No. It reads and measures the source without writing, re-encoding, trimming, or removing frames or audio.
What's the difference between an exact hash match and a near-duplicate?
An exact match means the tool generated the same quantized pixel checksum for sampled frames. A near-duplicate means the frames are visually similar by perceptual-hash distance, but not pixel-identical. The report labels the match type.
Why might it miss a fast cut or short black flash?
Detection only evaluates frames at the chosen sample interval. Lower Sample interval (ms) for closer temporal coverage, with more analysis work.
What if my video has no audio track?
The silence panel shows No decodable audio stream was found in this file. instead of creating silence events.