Loading page
About this tool
Sample bounded frames from a local video and review near-duplicate runs or high visual changes without uploading the media or making an authenticity claim.
The Video Frame Continuity Inspector samples up to 60 frames from a local video at an interval you choose, compares each adjacent pair by mean RGB pixel difference and 16-bin luminance histogram distance, and flags the runs that read as near-duplicate or high-change. It is a sparse visual-change screen for editors, reviewers and researchers who want a defensible starting point for where to look in a clip — not edit detection, deepfake detection or an authenticity claim, and the exported report says so in its own limitations list. Everything decodes and compares in your browser; the video is never uploaded.
Open Video Frame Continuity Inspector on AltFTool — it loads instantly in your browser.
Press 'Choose a local video' and pick an MP4, WebM, Ogg, MOV or M4V up to 80 MB and 10 minutes; duration and dimensions are checked before sampling.
Set the 'Sampling interval' slider between 0.25 and 30 seconds, then press 'Run continuity screen' — the plan is capped at 60 sampled frames.
Review the near-duplicate and high-change cue counts on the sample-to-sample timeline and press 'Download JSON' for video-continuity-counts-and-timing.json.
Each pair gets both a per-channel pixel difference and a 16-bin luminance histogram distance, so a lighting shift and a content shift are distinguishable.
Consecutive pairs carrying the same cue collapse into a single start/end range with its peak values, so you get a handful of intervals to review rather than 59 rows.
The exported JSON carries the schema version, sample counts, cue ranges and an explicit limitations list, so the screen cannot be quoted as a verdict it never made.
A pair is near-duplicate when the normalised pixel difference is at or below 0.008 and the histogram distance is at or below 0.012; it is high-change when the pixel difference reaches 0.18 or the histogram distance reaches 0.28. Anything in between is recorded as an ordinary change.
MP4, WebM, Ogg, QuickTime MOV and M4V files up to 80 MB and 10 minutes, with source dimensions no larger than 4,096 pixels on an edge. Frames are downscaled to a working copy of at most 320 pixels on the long edge before comparison, which keeps the pass fast and memory-bounded.
No. It is a screening aid only. Cuts, flashes, fades, camera motion, lighting changes, graphics and even codec or browser decoding differences all produce high-change cues, while static scenes and coarse compression produce near-duplicate cues without any editing freeze — and only the sampled instants are inspected, so an absence of cues proves nothing.
The interval ranges from 0.25 to 30 seconds and the pass is capped at 60 sampled frames, so a 10-minute video needs roughly a 10-second interval to cover the whole runtime. Thumbnail previews are shown for up to 24 sampled frames; beyond that the numeric results still run.