Why Does a Subtitle Appear One Frame Early or Late After I Convert Its Timing Format?

Milliseconds and frame positions are different time grids

WebVTT specifies cue timestamps using thousandths of a second. A frame-based caption workflow represents positions relative to the video's frame rate. Those grids do not always coincide.

At a constant 25 frames per second, frame boundaries are 40 milliseconds apart. A cue intended to start at 1.019 seconds lies between the boundaries at 1.000 and 1.040 seconds. A conversion that requires a frame boundary must choose how to represent that time.

Rounding to the nearest boundary gives a different result from always rounding upward. Neither choice automatically proves the caption is correctly synchronized to the actual speech. The editorial goal is to place the cue appropriately for the final media and destination format.

First confirm the exact source timestamp and the actual frame rate. Do not infer the rate from a project nickname or assume that a nominal decimal rate and its exact fractional counterpart are interchangeable.

Identify the conversion that introduced the difference

Preserve the original caption file and create a separately named converted copy. Compare cue starts and ends before and after conversion. Record whether the destination format uses milliseconds, frames or another time representation.

For each suspect cue, calculate its position relative to nearby frame boundaries. If a one-frame difference appears only after conversion, inspect the conversion policy. If the source cue was already late relative to speech, the problem predates the format change.

Repeated conversion can compound uncertainty. Converting milliseconds to frames, then back to milliseconds, may not reproduce the original timestamp exactly. Avoid using a previously rounded derivative as the authoritative timing master when the original high-precision cue record is available.

Also inspect cue endings. A rounded end may collide with the next caption or create an unwanted gap even when the start looks correct. The final player's behavior matters more than an abstract claim that every timestamp must remain numerically identical.

Hypothetical case: a cue starts between two 25 fps frames

Hypothetical calculation; no subtitle converter or player was tested. A fictional product video is constant 25 fps. Its source WebVTT cue begins at 00:00:01.019. The neighboring frame boundaries are 1.000 seconds and 1.040 seconds.

Under a nearest-boundary rule, the cue would begin at 1.000 seconds, 19 milliseconds earlier than the source timestamp. Under an upward-boundary rule, it would begin at 1.040 seconds, 21 milliseconds later. Both are less than one 40-millisecond frame interval from the original.

Suppose the merchant reports that the caption appears one frame late. The editor should inspect which conversion rule was actually applied and whether the caption still matches the spoken word. The calculation alone does not establish that the later placement is wrong.

The worksheet records the two mathematical alternatives. It does not claim that any named caption converter implements either rule by default.

Review the boundary against real speech

Listen to the final audio and inspect the caption at normal speed, then step through the relevant boundary. Check the first meaningful word, the final word and any nearby cue transition.

If the discrepancy is caused by conversion precision, correct the supported export or timing workflow and retain the original master. If the discrepancy comes from a trimmed or retimed video, use the actual edit mapping instead. Do not apply a global one-frame shift without checking unaffected cues.

The existing frame-rate review guide explains why reviewers may identify different moments with the same frame number. This article isolates the numerical rounding introduced when caption timestamps are represented on a frame grid.

Record your own observations

The filled rows are illustrative examples. Use the blank row for your own files and observations.

Download the editable CSV worksheet

What the references establish

• W3C — WebVTT: The Web Video Text Tracks Format: Primary specification for WebVTT timestamps with three fractional-second digits and start/end cue timing.
• Adobe Premiere — Create captions: Documents caption formats and timing controls in a video-editing workflow; does not establish a universal rounding policy for format converters.