There are two ways subtitles go out of sync, and they look similar for the first minute of playback. One is a constant offset: every cue is late by the same amount, so the subtitles are wrong from the very first line. The other is drift: the subtitles are right at the start and progressively worse as the video plays. People describe both as "out of sync" and reach for the same tool, then wonder why shifting by a fixed number of milliseconds did not help.
The distinction matters because a constant offset has a single correct fix and drift does not. Shifting a drifting file just moves the point where it is correct. This guide is about recognising which one you have, measuring it, and choosing a repair that actually holds.
Telling the two apart in thirty seconds
Open the video, jump to a line of dialogue near the beginning, and note whether the subtitle appears before, on, or after the sound. Then jump to the last few minutes and do the same. Three outcomes:
- Wrong at the start, wrong by the same amount at the end — constant offset. Every cue needs the same shift.
- Correct at the start, wrong at the end — drift. The file needs to be rescaled, not shifted.
- Wrong at the start by one amount, wrong at the end by a different amount — drift plus offset. Both corrections are needed, and the offset has to be fixed first.
A useful sanity check: if shifting by the amount you measured at the beginning fixes the beginning and makes the end worse, you have drift. That is the single most common reason people conclude "the shift tool does not work".
Where drift comes from
Drift is a difference in elapsed time between the file and the video. Something made one of them run at a slightly different rate than the other, and the error accumulates. In practice it comes from four places.
1. Frame rate conversion
The most common cause by a wide margin. Film runs at 23.976 frames per second; European television runs at 25. A 24 fps master conformed to 25 fps plays about 4.2% faster, so a 90-minute film finishes roughly 3 minutes 45 seconds early. Subtitles timed against the original will be correct at the start and nearly four minutes ahead by the credits.
The reverse happens too: 25 fps slowed to 23.976 fps makes the video 4.3% longer, and subtitles fall progressively behind. If your drift is close to 4%, frame rate conversion is almost certainly the explanation, and the correction is a clean ratio rather than a guess.
2. Variable frame rate and broken timestamps
Some screen recordings and phone videos report an average frame rate while actually varying it. Subtitle timing is derived from frame numbers times that average, so the accumulated time drifts against real elapsed time. This is harder to correct because the error is not linear, though a linear approximation often gets close enough for viewing.
3. Re-encoded or edited video
If scenes were cut, an intro was removed, or a "previously on" recap was added, the video timeline changed but the subtitle file did not. That produces a step change at the edit point rather than smooth drift: correct, then suddenly wrong by a fixed amount. No single global correction can fix it, because the correct offset is different before and after the cut.
4. Subtitles authored against the wrong source
Subtitles for a theatrical release used with a home video version that has a different studio logo at the front. The result looks like a constant offset — until the first scene change, where it may shift again. Worth ruling out before doing any maths.
Measuring drift precisely
Two measurements are enough. Find a cue near the start with an unmistakable audio marker — a door slam, a name spoken, a gunshot — and note the difference between when the subtitle shows and when the sound happens. Then do the same with a cue near the end. Write both down in milliseconds.
| Position | Video time | Subtitle shows | Error |
|---|---|---|---|
| First cue | 00:01:12 | 00:01:12.400 | +400 ms late |
| Last cue | 01:28:30 | 01:29:03.100 | +33 100 ms late |
The error grew by 32 700 ms over 5 238 seconds of video. That is 0.624%, which is close to the 4.2% of a 24→25 conversion but not equal to it — so this file has drift from a re-encode rather than a clean frame rate change. The number you want is the ratio, not the total.
Rescaling: the fix that actually works
Drift correction is a linear rescale: multiply every timestamp by a factor and add an offset. In the example above, the factor is 1.00624 — the video is 0.624% longer than the file assumes — and the offset pulls the whole thing back by 400 ms to fix the start.
new_time = old_time × 1.00624 − 400 ms
For frame rate conversions the factor is exact and worth memorising:
| Conversion | Factor | Effect on a 90-minute file |
|---|---|---|
| 23.976 → 25 fps | 0.95904 | 3 min 45 s early |
| 25 → 23.976 fps | 1.04271 | 3 min 51 s late |
| 24 → 25 fps | 0.96 | 3 min 36 s early |
| 25 → 24 fps | 1.04167 | 3 min 45 s late |
Note that 23.976 and 24 are not interchangeable. The 0.1% difference sounds negligible and produces about five seconds of drift over ninety minutes, which is enough to be visible.
When rescaling is the wrong tool
Rescaling assumes the error grows smoothly from zero. Three cases break that assumption:
- Cuts and inserts. If the video was edited, the error jumps at the edit point. Find the jump, split the file there, and treat each part separately.
- Chapter-by-chapter timing. Some releases are timed per chapter with their own small errors. Rescale each chapter rather than the whole file.
- Files with cues outside the video duration. If the last cue is at 2:10 and the video is 2:00, you are probably timing against a different cut, not drifting.
In all three cases, check the file against the video at three or four points rather than two. If the error does not grow steadily, it is not drift.
A procedure that works
- Measure the error at the first cue and the last cue. Write both numbers down.
- Subtract one from the other and divide by the elapsed video time between them. That is the drift rate.
- If the rate is within 0.05% of a known frame rate ratio, use the exact factor instead of your measurement — measurement error is larger than the difference.
- Apply the scale, then apply the constant offset needed to fix the first cue.
- Re-check at the start, middle and end. If the middle is worse than both ends, the file is not linearly drifting and needs splitting.
Most people stop after step four and get a file that is good enough. Step five is what separates a fix from an approximate fix.
Avoiding the problem next time
If you are producing subtitles rather than repairing them, the practical safeguard is to record which video version they were timed against — resolution, frame rate, and where it came from. A line in a text file next to the subtitles is enough. The expensive part of fixing drift is not the arithmetic, it is reconstructing what the original source was.
If you are repairing a file, keep the original. A rescale is lossy in the sense that rounding errors accumulate across repeated operations, and it is common to apply one, decide the factor was wrong, and apply another. Working from the pristine file each time avoids compounding the error.
Try it on your own file — no upload, nothing leaves your browser.
Open the Subtitle Time Shifter - Fix Out-of-Sync SRT/VTT Subtitles →