Subtitle Workbench
← All guidesWorked reference · Updated 11 October 2026

A fixed subtitle delay and subtitle drift need different adjustments

Measure two anchors before deciding whether an offset or a linear retiming is appropriate.

Measure a beginning and an ending

Choose a clearly identifiable word or sound near the beginning of the video, then another near the end. Record where each occurs in the subtitle timeline and in the final video. Avoid a vague pause or a scene transition whose exact instant is hard to agree on.

If both subtitles are late by roughly the same amount, a fixed offset may solve the mismatch. If the gap becomes larger over time, changing only the offset will align one section while leaving another section wrong. That pattern is drift and may need a timing scale, an edit-specific repair or a different subtitle source.

Fixed offset: shift every retained timestamp

new_time = original_time + offset
+1500 ms: subtitles display 1.5 seconds later
-1500 ms: subtitles display 1.5 seconds earlier

The sign matters. A positive adjustment delays the subtitle. A negative adjustment makes it appear earlier. Preview a few cues after changing the setting; do not infer the direction from the word 'sync' alone.

If an earlier shift would place a cue before zero, the workbench refuses the export by default. The alternative policies are explicit: trim the portion before zero, or drop a cue whose start becomes negative. Both can lose information, so inspect the change report before downloading.

Two-anchor linear retiming

scale = (target_B - target_A) / (source_B - source_A)
offset = target_A - source_A * scale
new_time = original_time * scale + offset

For example, source anchors at 10 and 110 seconds and target anchors at 12 and 114 seconds produce a scale of 1.02 and an offset of 1.8 seconds. The workbench applies this same transformation to cue starts and ends, then rounds to milliseconds.

Both pairs must increase and use the same unit. The interface accepts seconds for anchors and displays the resulting offset in milliseconds. Reversed or equal anchors are rejected. A very large scale or offset is also rejected if it exceeds the safe numeric range.

When a linear adjustment is inappropriate

A video with removed scenes, inserted introductions or several independently edited sections will not necessarily have one consistent timing scale. Align each section with evidence from the final cut. Retiming also cannot repair incorrect words or an unfinished transcription.

The two-anchor calculation is deterministic, but the choice of anchors remains a human judgment. Inspect a middle point after previewing the endpoints. Export a new file and retain the input as a reference.

Try the related tool