AI dubbing on the same caption timeline | AI Video Factory

Dubbing usually fails for a boring reason: you export captions into another app, lose the clock, and every rewrite means a full re-record. In the studio the calmer path is: review the cue list, then synthesize each line on the same timeline. Dubbing is written as an extra track; the original audio can stay.

Clean cues first

Copy with no timestamps is read end to end, and length is a guess. The order that holds:

1. Build a timestamped cue list (speech, burned-in OCR, or an imported SRT).
2. Fix typos, names and breaths, then start dubbing.
3. Synthesis reads the current cues — do not open a second project.

The ten-minute caption path is in this guide. Dubbing a messy axis only scales the mistakes.

Dubbing on the same timeline

1. Open the studio and continue a project or drop in a local MP4 / MOV. The file stays in the browser.
2. Confirm the cue list is reviewed. If you have none yet, recognize speech or import an SRT first.
3. Tick AI dubbing. Chinese uses 晓晴; English picks from several voices. With auto-detect, Chinese lines use 晓晴 and English lines use the English voice you chose.
4. Speed follows each cue’s duration by default. If one line sounds thin and rushed, change that line’s words or in/out — do not drag the whole file to 1.4×.
5. Keep original audio on by default. Dubbing is an extra track and the picture stays locked to the source. Mute the original only when the voiceover should fully replace it.

The free plan includes one dubbing run per month. Voices, mix and limits are on Features and Pricing.

Review only three things before you generate

• Punctuation: quotes and ellipses cause odd pauses — turn them into full stops or commas.
• One breath per cue: two sentences in one cue make the voice inhale in the wrong place.
• Mixed Chinese and English: brand names can stay English, but if whole clauses switch language, auto-detect with two voices is stabler than locking one.

What usually breaks

Treating dubbing as “read the whole file”: empty heads, laughs and stingers get spoken. Delete or empty those cues first. Always listen through — the model is for speed, you own mouth shape and level.

If you still need a translation, translate first, then dub; the clock does not move. The full order is in creator post in 2026.