Out-of-sync video? SyncProof finds the delay and fixes it.

Offset is a fixed delay: the audio is, say, 120 ms late from the first second to the last. Drift is a delay that changes: the video starts in sync and slowly slides out, so it looks fine at minute 1 and clearly off at minute 40.

Drift usually comes from two devices with slightly different clocks (a camera and a separate recorder, or a PC recording a stream), or from variable-frame-rate video being converted. SyncProof measures sections across the whole video and fits a drift line through the readings, reported in milliseconds per minute.

Pro corrects a steady offset, a steady drift, and sync that jumps part-way through, for example after dropped frames or a recorder glitch. Each part gets its own correction, the joins go at a cut or a pause in the speech, and every part is measured again after the fix. If a part has too few clear readings to fix safely, SyncProof tells you instead of guessing. Jumps smaller than about 25 ms can't be told apart from normal variation, so they aren't split out.

Videos where someone speaks on camera, in MP4, MOV, MKV and most other formats FFmpeg can open. A separately recorded soundtrack can be WAV or most other audio formats, at any sample rate. For a reading, SyncProof needs:

• a face in view, roughly facing the camera (turned less than about 50°). Small faces in wide shots work, down to about 4% of the frame height;
• speech you can hear. Music or effects under the voice lower the confidence, and the report shows that.

Shots shorter than 1.5 seconds are skipped. Voice-over with no visible speaker, animation and back-of-head shots can't be measured, and SyncProof says so instead of guessing.

In our tests we shifted the audio of 8 clips by 0, ±40, ±80 and ±200 ms. All 56 runs passed: the measured offset followed the shift within 2.3 ms at worst, and 0.3 ms on average. One frame is 33–42 ms at common frame rates.

The absolute zero point has about ±25 ms of uncertainty (under one frame), because lips naturally move slightly before the sound. Each reading comes with a confidence level. On our calibration footage, readings marked "good" were within ±35 ms 99% of the time, and "fair" ones 94% of the time.

It isn't perfect, and it doesn't pretend to be: when there's no usable face or speech, it gives no answer rather than a wrong one.

Up to 10 minutes, SyncProof analyses every frame. In our tests on a 4-core CPU that ran 1.6 to 5 times faster than real time, so a 10-minute video takes about 2 to 6 minutes. The slow end was 1080p with four faces in every frame: more faces means more work. No graphics card is needed.

Longer videos are sampled: about 20 to 48 short stretches spread over the file, plus extra ones wherever the sync changes. In our tests a 30-minute video took under a minute to check, and about 2½ minutes to fix and re-check including finding where its sync jumped. A 2-hour podcast took about 3 minutes to check. Sampling can miss a problem shorter than the gap between samples (about 3 minutes on a 2-hour file); you can ask it to analyse every frame instead.

No. SyncProof doesn't generate or edit mouth movement, unlike AI "lip-sync" tools. It measures the delay and moves the audio. The picture is copied as it is, without re-encoding.

No. Checking and fixing both run on your PC, and your footage doesn't leave your computer. See the privacy policy.

SyncProof looks at every face, ignores the ones whose mouths aren't moving, and uses the face whose lips best match the speech. In our multicam test with four people talking in a wide shot, it picked the actual speaker in every wide-shot section and found all 9 cuts. If two people talk at the same time, it uses the stronger match.

Within one video file, SyncProof searches up to half a second (500 ms) early or late. A separate recording (a field recorder, Zoom or Riverside track) can start minutes before or after the camera: SyncProof first lines it up with the camera's own audio, including a slightly different clock, then checks the lips. That needs the camera to have recorded some sound of its own.

SyncProof measures what's there. In dubbed or studio re-recorded (ADR) dialogue, sync can vary from line to line, and the section-by-section graph shows that.

One person, on up to two PCs (for example a desktop and a laptop). You can use it for your own videos and for client work. The files you fix are yours.

Yes. If Pro doesn't work for you, ask within 14 days of purchase for a full refund. Reply to your receipt email or write to the support address below. The free check lets you see what SyncProof reads in your footage before you buy.

Not yet. SyncProof runs on Windows 10 and 11. We may make a Mac version later, but there's no date, and we won't take payment for one before it exists.