18+ · public-domain clip index

Reverse clip & image search

Drop a screenshot, GIF, or short clip. SourceMatch fingerprints it on your device and looks for the source frame it came from, in an index of public-domain films & newsreels from the Internet Archive. Verified hits are Match (with the proof); close-but-unverified hits are Similar (a lead, clearly labeled).

How a clip becomes a search

  1. Pick the frames that count

    Your browser samples the clip at even points (images: the frame itself) and keeps up to 24.

  2. Fingerprint on your device

    Each frame becomes a compact perceptual fingerprint — a 64-bit structural hash plus ~24 local feature descriptors — right in your browser. The original file never leaves your device.

  3. Gate, then verify

    The hash narrows the index to a handful of candidates. Each candidate is then checked geometrically: one transform (scale + rotation + translation) must fit the matched features. Verified hits are labeled Match with a confidence score and the proof; unverified ones are labeled Similar as a lead.

Match

A verified same-picture result. Includes a 0–100 confidence and the frame pair that proves it — matched features are drawn on both images.

Similar

Looks alike to the fingerprint but did not pass geometric verification — shown as a lead, never as a confirmed answer.

Common questions

Is my file uploaded?

No. Your file is read locally in your browser. Only a small fingerprint — not the file and not a full image — is used for the search. Candidate index data is fetched from our own CDN.

What’s the difference between a Match and a Similar?

A Match is a verified result: the matched features line up under a single geometric transform, and the confidence reflects how tightly. A Similar is an unverified lead — close on the structural hash but not geometrically confirmed — and is clearly labeled as such.

Is this a real search? What is in the index?

Yes — the built-in index is a set of public-domain films and newsreels from the Internet Archive (Prelinger collection); every result links back to its source item and its public-domain status. The search finds what is in that index; to find arbitrary clips from anywhere on the internet, the index would have to lawfully contain them, which this build does not attempt. The “QA test patterns” shown with results are a separate synthetic set used to verify the pipeline itself.

Why don’t I get a match for my clip?

Reverse search only finds content that was previously indexed. If your clip isn’t from one of the indexed public-domain films (or a near-identical re-encode/crop of one), there is nothing to match against — the honest answer is “no results”, not a guess.

How do I report a problem or ask to remove content?

Use the removal request link. We review removal requests promptly. (The index is public-domain material; requests typically concern how an item is presented.)

What about ads?

Advertising is off by default in this build. When enabled, ads are clearly labeled “Advertisement,” non-intrusive, and never intercept your actions. See the privacy note.