Near-duplicate clip
Also known as: near duplicate, repeated take
Two clips that are not identical files but show essentially the same thing, which is what makes an automatic edit read as a loop.
Updated
What it means
A near-duplicate is the same moment twice over, from a slightly different angle or on a later attempt. Nothing about the files matches, so no checksum will catch it, yet a viewer plainly sees one event repeated.
Six laps of the same park feature are six near-duplicates. An edit that keeps all of them stops looking like an edit and starts looking like an unfiltered dump of the card.
This is the single biggest reason automatic highlights disappoint. Each repeat scores well on its own merits, so a Highlight detection model ranking moments independently cheerfully keeps every one of them.
How it works in practice
Detection compares meaning rather than bytes. Each clip is reduced to a numerical description and clips whose descriptions sit close together are treated as the same event, which is what Video embeddings are for.
Having found a group, the pipeline keeps the strongest member and discards the rest, so the decision depends on a reliable ranking from Shot scoring rather than on which file happened to come first.
Grouping before selection is the ordering that matters. Filtering duplicates after the cut is assembled means rebuilding it, while filtering before means the selection never sees the repeats.
What to watch out for
The threshold is genuinely hard to set. Loose, and the second and third attempt vanish even though the third was the one that worked; tight, and the loop survives into the finished video.
Some repetition is also the point. A trick filmed from three mounts is deliberate coverage, not duplication, and collapsing it to one angle throws away the variety Multicam editing exists to exploit.
Long clips defeat whole file comparison. Forty minutes containing one repeated feature is not a duplicate of anything as a file, so the comparison has to run on segments to find anything at all.
How RawClip handles it
Clips showing the same moment are grouped and only the strongest survives into the cut, which is why a session of repeated laps comes back as a sequence rather than as the same feature over and over again.