Fixing subtitle drift: 23.976, 24, 25 and 29.97 explained
Your subtitles are fine for the first minute and two minutes late by the end. Nothing is broken: the file was timed against a different speed of the same film. This guide shows how to recognise that, how to compute the exact correction, and how to verify it without scrubbing the whole timeline by feel.
What drift actually is
A video has a frame rate: the number of still pictures it shows each second. A subtitle file has no frame rate at all. It stores timestamps in hours, minutes, seconds and milliseconds, which describe positions in real time. The connection between the two is made when the file is authored: someone watched a specific version of the video and wrote down the moment each line should appear in that version's clock.
If the version you are delivering is a different speed, every timestamp is wrong by a growing amount. A film transferred from 24 fps to 25 fps runs 4 percent faster, so its 60 minute runtime becomes 57 minutes 36 seconds. A cue that belongs at 30:00 in the original happens at 28:48 in the faster version. At the start, the difference is a few seconds; at the end, it is over a minute. That linear growth is what makes drift easy to diagnose and easy to fix: the error is proportional to the timestamp, not random.
The rates you will actually meet
Film is shot at 24 frames per second, but most digital film releases are stored at 23.976, which is 24000 divided by 1001 (23.976023976...). The awkward number exists because of the conversion from NTSC colour video, and it has followed us into streaming. Television in North America and Japan uses 29.97 (30000/1001), often written as 30 for short. Most of Europe, much of Asia and PAL DVD use 25. Web video is usually 24, 25, 30, 50 or 60, and phone footage is commonly 30 or 60.
Two pairs cause almost all the pain. The first is 23.976 against 24: a difference of 0.1 percent, which is 3.6 seconds per hour. It is small enough to survive a quick review and large enough to be visible at the end of a feature. The second is 23.976 against 25: a difference of 4.27 percent, which is 2 minutes 34 seconds per hour. A PAL DVD of a film timed for its 23.976 master will be dramatically out.
The exact factors
The correction is a single multiplication. If the subtitles were timed for a video running at F1 and your video runs at F2, multiply every timestamp by F1 divided by F2. That is all the maths there is, but the direction is easy to get backwards, so check one known cue before applying it to the whole file.
| Timed for | Your video | Multiply by | Error after 1 hour | Typical cause |
|---|---|---|---|---|
| 23.976 | 24 | 0.999000 | -3.6 s | Film master vs 24 fps export |
| 23.976 | 25 | 0.959041 | -2 min 27.5 s | PAL version of a film |
| 24 | 25 | 0.960000 | -2 min 24 s | PAL speedup |
| 25 | 23.976 | 1.042709 | +2 min 33.8 s | Film master from a PAL file |
| 24 | 23.976 | 1.001001 | +3.6 s | 24 fps timing on a 23.976 master |
| 29.97 | 30 | 0.999000 | -3.6 s | NTSC timing on a 30 fps export |
| 30 | 29.97 | 1.001001 | +3.6 s | 30 fps timing on an NTSC master |
Notice that the factor is below 1 when your video is faster than the file expects. A faster video reaches every moment sooner, so the timestamps must shrink. If the last cue is late, its timestamp is too large, so you need a factor below 1. If the last cue is early, the timestamp is too small and the factor is above 1. Checking the direction on the final cue is the fastest sanity test you can do.
Telling drift from a constant offset
A constant offset means every cue is equally early or late. It usually comes from a missing logo, an added recap, a different distributor card, or a cut that was trimmed at the head. Drift means the error grows. The test takes two cues:
- Find a hard sync point in the first minute: a door slam, a cut, a clap, the first word after silence. Note the error.
- Find a hard sync point in the last five minutes. Note the error again.
- If the two errors are the same, use the offset tool. If the later error is larger, use the frame rate tool first, then a small offset.
Order matters. Scaling after an offset also scales the offset, because the transform is anchored at zero. If you add 12 seconds and then multiply by 1.0427, those 12 seconds become 12.5, and the first cue is half a second out. Do the scale first, check again, and trim the residual with the offset tool.
Applying the correction in CaptionSync
Open the file, then set the two frame rates in the timing panel: the rate the timing was authored for on the left, the rate your video plays at on the right. The panel shows the factor and what it does at the one hour mark before anything changes. Apply it, then look at the last cue: it should now sit within a frame or two of the picture. If it is uniformly late by, say, 800 milliseconds, use the offset tool for the remainder. If it is early, use a negative offset.
If you do not know the source rate, two point sync does the same job from evidence. Pick two cues whose correct times you know, type the current and target timecodes, and the tool fits a straight line through both points: the slope is the frame rate ratio and the intercept is the constant offset. It is what you would do by hand with a calculator, with fewer places to mistype a digit.
Verifying at both ends
A correction that looks right at the start can still be wrong at the end if the factor was slightly off, so check three points: the first cue, a cue near the middle, and the last cue. On a feature, an error of half a second at the end is normal after rounding; an error of five seconds means the rate pair was wrong. If the first cue is right, the middle is right and the end is late, the source might have been a different cut rather than a different speed, which no frame rate factor will fix.
Four mistakes worth avoiding
- Applying the factor twice. Rescaling an already corrected file doubles the error in the other direction. Undo is there for a reason: one pass, then verify.
- Rounding the factor to 1.04. A 4 percent guess leaves 22 seconds of drift on a feature. Use the exact ratio, 1.042709.
- Correcting before checking the master. Ask for the runtime and frame rate of the actual delivered file. The brief is often older than the export.
- Ignoring drop-frame timecode. 29.97 drop-frame changes how timecode is labelled for broadcast, not how subtitle timestamps relate to the picture. Do not apply a drop-frame conversion to a subtitle file.
Fix the drift in the studio
Open your file, pick the two rates and check the last cue. The factor is shown before anything moves.
Questions about frame rate drift
Why is 23.976 used instead of 24?
23.976 is 24000 divided by 1001, a rate inherited from the conversion between NTSC colour video and film. It is close enough to 24 for the eye and compatible with the older broadcast chain. The 0.1 percent difference accumulates to 3.6 seconds per hour, which is why a file timed for one rate can be visibly out on the other.
Does applying 25 to 23.976 break a file that is already correct?
Yes, it makes it wrong in the opposite direction: the cues would become 4.27 percent later. CaptionSync shows the factor and the effect at one hour before applying, and Undo restores the previous state exactly. Verify the last cue against the picture before you apply the factor to a whole feature.
Do I need to worry about drop-frame timecode?
Only for broadcast timecode labelling, not for subtitle timestamps. Drop-frame numbering skips frame numbers to keep 29.97 timecode roughly aligned with clock time; a subtitle file stores clock time, so the frame rate ratio is the only correction that applies.