A caption job is three clocks that people collapse into one complaint. Transcription is whether the words are right. Timing is whether they arrive with the mouth. Layout is whether they sit in a readable place. 'The captions are off' is usually timing: the line lands late, hangs after the speaker has moved on, or overlaps the next sentence because two cues share a frame.
Word-level timestamps are a finer clock than line-level ones. Karaoke / word-highlight families (Snap, Reveal in Caption Studio) need a timestamp per token so the accent can move with the voice. Ordinary subtitle families need a start and end per line. Both are alignment, not styling. Forced-alignment is the model of that clock: it maps a known transcript onto the audio. Auto captions run speech-to-text and alignment together, which is why a wrong word and a late word look like the same failure until you read the file.
A whole track that is uniformly early or late is an offset: trim an intro and every cue needs to move by the same number of seconds. A track that drifts further out as the video runs is a frame-rate mismatch (a 25 fps caption file against 23.976 fps picture is the usual case) and an offset will make the start right and the end worse. Burned-in captions freeze the clock into pixels; fixing timing after burn-in is a new render, not a nudge.
In practice
- Read the first and last thirty seconds before you accept a caption render — drift shows up at the end.
- Lock the cut before you caption. A later trim is a new timing pass.
- If only one name is late, that is a transcription/alignment miss on that word, not a global offset.
The mistake to avoid
Applying a constant offset to a track that is drifting. You will fix the hook and break the closer.
Where you will run into it
- Add Captions to a Video — Speech in, styled subtitles out — no manual timing.
- AI Caption Generator — Drop a video, get timed captions. Free, on your device, no upload.
Related terms
Burned-in captions
Burned-in captions meaning: subtitles rendered into pixels, so viewers cannot turn them off or lose them on re-upload. Closed captions are a separate track.
Word-highlight captions
Word-highlight captions are burned-in subtitles that light up each word as it is spoken, using word-level timing and a caption-style accent colour.
Reading speed
Caption reading speed is how fast on-screen words arrive, in words per minute. A cue gone before it can be read is a timing error, not a style one.
Speech-to-text
Speech-to-text meaning: converting spoken audio into written text, the transcript captions, translation, and search all depend on.
Forced alignment
Forced alignment meaning: matching a known transcript to its audio to find exactly when each word was spoken.
The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync, in your browser or on your phone.
Free on the web and on iPhone, one account.