Few delivery defects feel as personal as sync drift. You watch the lip-sync — or the splash landing exactly on the sound — a dozen times in Premiere and it’s right. You know it’s right. Then the published video starts in sync and slowly wanders, until by the end the sound is arriving noticeably early or late and the whole piece feels drunk. Viewers won’t name the problem. They just feel the cheapness and leave.
Here’s the thing, though. Sync drift after upload is almost never the platform being malicious, and honestly it’s rarely a mystery at all. It’s usually baked into your file before upload by frame-rate handling, sometimes helped along by sample-rate sloppiness or an export setting nobody checked. And the encouraging part? Once you know where sync goes to die, the fixes are short and permanent.
We ship AI-generated dialogue and beat-synced animal shorts weekly, so this checklist is the one we actually run whenever a test upload’s tail stops matching its head. Every time.
Where Sync Goes To Die: VFR And Mismatched Rates
The number one cause of drift — by a wide margin in our experience, and in widely documented editor behaviour — is variable frame rate video walking into a constant frame rate world.
Phones, screen recorders, and some generation or capture tools record variable frame rate (VFR) footage. That means the time between frames stretches and shrinks to save space or cope with load. The file still reports a nominal rate like 30 fps, but individual frames aren’t evenly spaced. Meanwhile editors — Premiere included — and platform encoders strongly prefer constant frame rate (CFR), where every frame lasts exactly the same time.
So. When VFR meets a CFR timeline, something has to give. The editor maps frames onto its constant grid as best it can. A short clip may survive unnoticed. But across minutes? Those tiny mapping errors accumulate in one direction, and audio — which runs on its own steady clock — slowly separates from picture.
The classic signature: perfect sync at the head, a beat off in the middle, obviously wrong at the tail. A fixed offset is a different problem (usually a simple shift). Drift that grows is a timing-base problem. And VFR is its favourite source.
Mismatched rates cause the quieter version of the same failure:
- 23.976 vs 24, 29.97 vs 30, 59.94 vs 60. These near-identical rates differ by 0.1 percent — nothing for ten seconds, very audible across a few minutes if picture and audio get conformed at different rates somewhere in the chain.
- Sequence set to one rate, clips generated at another. Our Seedance and Veo generations arrive at whatever rate the tool produced. If the Premiere sequence assumes a different rate and interprets the footage instead of conforming it honestly, timing stretches. You know what I mean? It’s subtle until it isn’t.
- Mixed sources conformed carelessly. A generated clip at one rate, a screen recording at VFR, a voice track recorded separately — each can be fine alone and drift together if the timeline papers over their differences.
The Constant Frame Rate Rule
Our rule for anything entering a delivery timeline: constant frame rate in, constant frame rate out, one rate from generation to publication. No exceptions for “it played fine in the viewer.”
How we enforce it in practice:
- Know what your sources actually are. Check suspicious clips before editing. In Premiere, clip properties and the metadata panel reveal frame rate; tools like MediaInfo show plainly whether a file is constant or variable. Phone and screen-recorded material gets checked every time. Generated clips get checked per tool and per batch, because tools change defaults quietly. They do. Trust me on that one.
- Convert VFR to CFR before editing, not after. Transcode variable-rate sources to a constant rate that matches your sequence (HandBrake’s constant framerate option, FFmpeg, or Premiere/Media Encoder workflows all serve here). Converting before the edit means sync decisions, caption timing and music cuts are all made against honest timing. Converting after just preserves already-drifted decisions more cleanly.
- Set the sequence rate deliberately. For most short-form delivery we edit at the rate we generated or captured at, commonly 30 fps constant for vertical. The sequence rate is the project’s clock. Choose it once, on purpose, and conform everything to it.
- Interpret footage honestly. When Premiere’s Interpret Footage is used to force a clip to the sequence rate, understand you are stretching or squeezing its timing. For B-roll, often harmless. For synced dialogue or action-on-sound, that stretch is precisely how drift is born.
- Export CFR, always. The export must be constant frame rate at the sequence rate. Any export path that writes variable timing hands the platform a drifting file with your name on it.
This rule is boring, which is its strength.
Teams discover VFR exactly once per painful project and then start converting sources on ingest. Build it into your bin routine: raw generations in, checked, conformed, then edited. Our wider prompt-to-final-cut workflow treats source preparation as a real stage for this reason, not as housekeeping.
Sample Rate Mismatches Nobody Checks
Video people check frame rates and forget audio clocks. Audio runs on sample rate, and mismatches there drift too — more slowly, more sneakily.
The video world standard is 48 kHz. Music production and older defaults often run 44.1 kHz. Premiere handles mixed rates by resampling, usually correctly, but the chain gets fragile when several things combine: a voiceover recorded at 44.1 kHz, a music track at 44.1, a sequence set to 48 kHz, and an export where audio settings were left on some inherited preset. Resampling errors are tiny per second and additive per minute. And exported files occasionally carry the mismatch forward instead of resolving it, especially when files pass through additional converters or screen-record steps.
Our audio hygiene is short enough to actually follow:
- Sequence audio at 48 kHz. Set it when the project is created, not fixed later.
- Record voice at 48 kHz when the device or app lets you choose. If a tool only gives 44.1, convert the file to 48 kHz once, deliberately, before editing, and edit with the converted file.
- Convert music and SFX to the project rate on ingest when building a library you will reuse. One conversion in the library beats resampling surprises across fifty projects. Fifty! Think about how much time that saves.
- Export audio explicitly: AAC, 48 kHz, high bitrate (we use up to 320 kbps for finals). Do not let the export inherit “same as source” ambiguity on a mixed-rate timeline.
- Be suspicious of screen recordings and chat-app audio. Voice notes and call recordings arrive at odd rates and sometimes variable timing. Convert them on arrival like VFR video, because that is effectively what they are.
Editor Settings That Export Sync Problems
Sometimes the sources are clean and the timeline is honest, and the export still ships drift. The usual suspects live in the gap between sequence settings and export settings — where inherited presets go to hide.
Export frame rate not matching the sequence. A 29.97 sequence exported through a preset that says 30 (or the reverse) forces a conform across the whole file. Short pieces shrug; longer pieces drift or duplicate/drop frames rhythmically. Match export rate to sequence rate exactly, including the decimal. “Close enough” rates are how 0.1 percent errors get hired.
Presets inherited from other work. The 4K landscape preset reused for a vertical short. The broadcast preset with its own audio assumptions. The quick-export someone saved in a hurry. Presets are wonderful until they describe a different project. We keep named delivery presets per format (the same discipline as our export settings for short-form) and treat any unnamed export as a draft. Honestly, that’s saved us more times than I can count.
Nested sequences and rate-mixed comps. Nesting a sequence of one rate inside a sequence of another, or sending a mixed-rate edit through dynamic-link-style round trips, can bake conforming into the render where nobody reviews it. If a project needs mixed rates, conform at the clip level before nesting, and test the export’s tail sync specifically.
Effects that retime silently. Speed ramps, frame-interpolation effects and optical-flow-style retiming alter picture timing locally. Applied to a shot with synced sound, they create local drift that feels random. Keep retimed shots on muted generation audio and lay sound deliberately, or accept the effect only where sync is not being judged.
Look, sync breaks at handoffs — where one stage assumes the next shares its clock. A checklist naming sequence rate, export rate and audio rate kills most of this before export. Ours sits in the pre-publish pass in our AI video QC checklist.
Testing Sync At Head, Middle And Tail
You cannot fix drift you have not measured. And “it felt off near the end” is not a measurement.
Test sync like an editor — at three points, against events that are easy to judge:
- Pick sync anchors. A door slam, a splash, a clap, a word with a hard consonant on a visible mouth, a beat hit with a visual accent. Every project has a few. Note their timecodes in the timeline while editing, when you are there anyway.
- Check the exported file first, before upload, at head, middle and tail anchors. In sync at the head and drifting at the tail in the export means the problem is yours and recent: rates, conforming, export settings. Fix before uploading anything.
- Check the published version the same way via your private test upload. If the export was tight and the published version drifts, suspect the file the platform received (VFR that slipped through, odd timing metadata) and re-deliver a clean CFR export rather than re-editing.
- Estimate the drift. A few frames at the tail of a short piece is a nudge-level problem. A growing gap measured in beats or syllables is a rate problem. Match the fix to the size: offsets get shifted, rates get rebuilt. Shifting audio to fix a rate problem just moves the point where it is wrong.
- Listen on the phone speaker for the final check. Sync perception tightens on small speakers with the picture at arm’s length — which is to say, in the exact conditions your audience judges you.
Re-Delivery Without Re-Editing
The good news buried in most drift cases: the edit is usually right and the file is wrong, which means re-delivery — not re-editing — fixes it.
If the timeline is in sync but the export drifts: audit sequence rate against export rate and audio rate, correct the preset, re-export, and verify the tail anchor. Ten minutes. Do not touch a single cut. Most “mystery drift” dies at this step. I’m serious — most of it.
If one source clip is the culprit (a VFR phone shot inside an otherwise clean project): convert that clip to CFR properly, replace it in the timeline with the converted version in place, check edit points around it (conforming can shift them by a frame or two), then re-export. The edit survives; the clock gets replaced.
If the whole project inherited a bad rate decision: create a new sequence at the correct constant rate, copy the edit across, and re-check anchors at head, middle and tail before re-export. Tedious, yes. But still an afternoon’s annoyance rather than a re-edit, and the anchors tell you immediately whether the copy translated cleanly.
On the platform side: replace the video where the platform allows it, or delete and repost from the corrected file, using your private-test slot first. Retire the drifting file from your archive as a source, clearly renamed, so nobody cross-posts from it later by accident. The master rule applies to broken children too: fix the master chain, re-export the children, never patch a platform copy.
Then write down what caused it. One line in the project notes: “OBS screen capture was VFR; convert on ingest”. Drift causes repeat until the lesson moves into the ingest routine, at which point they simply stop occurring.
That has been our experience exactly: sync drift went from a recurring delivery fire to a checklist item we stopped thinking about. Not because platforms improved — because nothing variable-rate reaches our timelines anymore.
Frequently Asked Questions
Why Is My Audio In Sync In Premiere But Out Of Sync After Upload?
Usually the drift was already in your export, too small to notice on a short timeline watch, and the platform faithfully preserved it. The common causes are variable frame rate sources conformed by the editor, a sequence/export frame-rate mismatch (29.97 vs 30 is the classic), or mixed audio sample rates. Check your exported file at head, middle and tail sync points before blaming the upload.
What Is The Difference Between Sync Offset And Sync Drift?
An offset is wrong by the same amount everywhere; shift the audio once and it is fixed. Drift starts in sync and grows, so no single shift fixes it — the head or the tail will always be wrong. Drift means a timing-base problem, usually frame rate or sample rate, and the fix is converting sources to one constant frame rate and one sample rate, then re-exporting.
Does Converting VFR To CFR Really Fix Audio Drift?
Yes, when VFR is the cause — which it most often is with phone and screen-recorded footage. Convert the source to a constant frame rate matching your sequence before editing (HandBrake’s constant framerate mode or FFmpeg both work), edit with the converted file, and export CFR. Converting after editing preserves drifted timing decisions, so the order matters.
Should My Audio Be 44.1 kHz Or 48 kHz For Video?
Use 48 kHz end to end for video work: sequence, recordings where you can choose, and export (AAC 48 kHz). Convert 44.1 kHz music and voice files to 48 kHz once on ingest rather than letting mixed rates resample quietly across a project. Short clips forgive rate sloppiness; longer pieces accumulate it into audible drift.
One Clock To Rule The Edit
Sync is a promise your whole chain makes to the viewer: picture and sound agree about what time it is. Keep one constant frame rate and one sample rate from the first generated clip to the final export, test the head, middle and tail like you mean it, and drift stops being weather and becomes what it should have been all along — a setting you chose once and never think about again.