Lipsync after caption timing

Final audio first. Burns second. Time captions to the final audio after lipsync. Captioning a pre-lipsync scratch track guarantees drift. The mouth move changes timing. The breath changes. Tiny gaps open. Your karaoke captions, once perfect on the scratch VO, now arrive early or late and make a clean performance look careless.

This is one of those pipeline rules that feels pedantic until you ship a week of slightly wrong burns and watch retention wobble for reasons nobody can name in the stand-up.

Why scratch-track captioning fails

Editors caption early because stakeholders want to "see it with text." Scratch tracks are useful for pacing picture. They are hostile to final caption timing when lipsync will rewrite the audio bed. Lipsync tools align mouth motion to a target performance or retarget speech. Even small shifts break word-level caption cues.

Burned captions are less forgiving than sidecar captions. A soft sidecar can be regenerated quietly. A burned file needs a re-export. That is why the order matters more when you deliver social masters with open captions baked in.

The fix is not "never temp caption." The fix is to treat early captions as disposable visualization. Delete them before final. Rebuild from the locked audio. The companion post captions after lipsync, not before spells out the same sequence for teams that need a shareable rule.

If multiple people touch the timeline, write the rule on the project: "No final caption job until lipsync lock." Without the written rule, someone helpful will caption on Friday night and force a weekend redo.

A clean audio-to-caption sequence

1. Lock picture length and performance intent.
2. Generate or record the voice you actually want to ship.
3. Run lipsync against that final voice (AI lipsync).
4. Spot-check mouth cadence on a phone with sound on.
5. Only then generate captions from the final mix (AI caption generator).
6. Style captions, burn or sidecar as needed, export.

If you change a single sentence after step 5, re-run captions for that region at minimum. Partial updates beat full denial. Pretending a one-word change will not matter is how one-word disasters happen on product names.

Music ducking belongs before caption final too when the ducking changes where syllables feel stressed. Captions should follow the mix listeners hear, not an unducked VO stem that only editors use.

For multilingual deliveries, still wait for lipsync lock on each language version. Do not reuse English caption timing on a Spanish lipsync and hope. Timing is language-specific even when the picture is shared.

Review habits that catch drift early

Watch once with sound off and captions on. Watch once with sound on and captions on. The mute pass catches readability. The sound pass catches sync. Many teams only do one pass and ship a confident mistake.

Zoom into plosives and names. Brand names and numbers are where timing errors look most embarrassing. If "2026" appears a beat late, viewers feel it even if they cannot articulate why.

Keep caption styles consistent across a series so reviewers judge sync, not whether this week used a new font bounce. Style experiments deserve their own test flight, separate from lipsync validation.

When stakeholders insist on seeing captions in an early cut, give them temporary burnt-ins on a copy labeled TEMP. Never overwrite the master timeline's final caption track with temp work.

Credits, rerolls, and calm production

Lipsync rerolls are sometimes necessary. Caption rerolls are cheap compared with regenerating a whole talking-head plate because text drift made the piece feel wrong. Protect the expensive stage by ordering the cheap stage correctly.

Versely costs are in credits for generation tools; human rework time is the hidden second currency. A correct sequence saves both. Put "lipsync lock" as a checkbox before "caption burn" in your template.

If you are assembling that template now, sign up for Versely and keep lipsync and caption tools in one project flow so the order is harder to scramble. The feeling to aim for is quiet competence: mouths and words agreeing without a heroic fix at midnight.

Teams that learn this once rarely unlearn it. Teams that ignore it keep hosting the same postmortem with new filenames.

Common failure stories worth avoiding

A team captions a brilliant scratch VO on Monday, lipsyncs a tighter performance on Wednesday, and ships Thursday without rebuilding text. Comments arrive about "weird captions." The editor swears the captions were perfect. Both are true in different timelines. Sync the timelines.

Another failure: someone updates one product name in the VO after legal review and patches audio surgically, then forgets captions still show the old name for two beats. That bug is tiny and brand-damaging. After any VO patch, search the caption file for the changed tokens.

A third failure: mixing burned English captions with a localized lipsync track for a second market. Picture stays, language changes, burns lie. Keep language-specific caption masters named clearly.

Build a lightweight gate in review: "Play 15 seconds mid-video with eyes closed, then eyes open on captions." If your ears and eyes disagree, fail the gate. No vibes debate required.

Also align freelancer incentives. If freelancers are paid for speed of first caption delivery, they will caption too early. Pay for final-safe captions after lipsync lock, or accept temp captions as non-billable drafts. Money teaches process faster than reminders.

If you deliver both a clean master without burns and a social master with burns, generate captions once from final audio and branch only the styling. Dual pipelines that caption independently will diverge. One caption source of truth is enough.

Tooling notes without magical thinking

Lipsync quality depends on clean audio, sensible frame rate, and a face that can carry the motion. Garbage VO produces garbage mouths, and captions cannot hide that. Fix voice performance first. Then lipsync. Then captions. Skipping to captions because text feels controllable is how teams decorate a broken bed.

When you use automatic captioning, still spot-check homophones and brand spellings after lipsync. Automatic systems love confident wrong words. A human pass on names and numbers is mandatory for client work.

If you change caption style presets mid-series, do it on a newly captioned final, not by restyling an old timing file that was built on scratch audio. Restyling can hide timing debt until someone watches with sound.

Ship checklist line worth tattooing on the template: final audio hash or filename recorded next to caption filename. If they do not match a known pair, rebuild. This is dull paperwork that prevents loud emergencies.

FAQ

Can I caption before lipsync if I use sidecar captions?

You can draft early, but rebuild before ship. Sidecars make fixes easier, yet shipping scratch timing still makes you look out of sync.

What if lipsync only changes a tiny region?

Re-time captions for that region and a small pad around it. Spot-check the boundaries where old and new timing meet.

Should music beds be locked before captions too?

Lock the mix listeners hear, including ducks. Captions should match the final experience, not an editor-only stem.