Check the clip before the story moves.
SyncLens gives a newsroom a fast forensic read on a video, with the uncertainty stated plainly, so a desk can decide how much weight to put on it.
The scenario
A video of a public figure saying something inflammatory is circulating. It might be genuine. It might be a voice clone over old footage. The story is live either way, and the desk has minutes.
Publishing a fake as real is a correction and a credibility hit. Ignoring a real one is a missed story. Both cost something.
The problem, on deadline
A yes-or-no verdict from a tool is not something a newsroom can print, because it hides how sure the tool actually is.
What a desk can use is a read that says which track looks manipulated, where, and how confident that is, so an editor can weigh it against everything else they know.
Which result matters most here
Real video with fake audio is the common shape of a fabricated quote: genuine footage with a synthetic voice reading a script the person never said.
Fake video with real audio and fully synthetic clips matter too. The per-segment timeline shows whether a whole clip is suspect or only the line being quoted.
- Read first
- Real video, fake audio
- Also relevant
- Fake video, real audio, Fake video, fake audio
Reading the report
The report is meant to inform an editorial judgement, not replace it.
1. Result and confidence. Plain language, with the confidence band first. A low-confidence read is a reason to keep reporting, not to publish.
2. The timeline. Whether the whole clip is in question or only the seconds being quoted.
3. The evidence. Tagged frames and the analysed audio, so a claim in the story can point at something specific.
4. Export. A PDF of the report to file with the story notes.
For verified newsrooms
Verified newsrooms and fact-checking organisations get a free tier. In return we ask to reference the work in a case study.
Export each report as a PDF to keep with the story.
Read the known limitations