When the sound in an AI video does not line up with the picture, the cause is almost never the model's audio inside a single clip. It is what happens after: clips joined without their sound and a voiceover laid back on top, captions timed from an estimate rather than the speech, a track that was trimmed or padded to fit a model's length, or an editor that treats a 24 fps clip differently from your other footage. Each one has a direct fix.
TL;DR
- Joining clips with Montage removes their sound. If you re-add the voiceover afterwards, the cuts have moved, so the voice has to be re-timed.
- Generated captions are an estimate. The Video Editor times them at about 160 words per minute, not from the voice.
- Some models cut or pad your track. Creatify Boreal uses only the first part of it; Hailuo H3's lip sync stops at about 15 seconds; Clips trims music over 15 seconds.
- AI clips are 24 fps. If you mix them with footage at another frame rate in your own editor, check frame rates first when the sound starts to drift.
- 8frame cannot re-sync lips on finished footage. Regenerate the line, or make a lip sync clip from a still and the audio.
1. The clips were joined, then the sound was added back
The Montage node joins 2 to 10 clips, and the merged file has no audio track: each clip's own sound, including generated dialogue and lip sync, is dropped. Montage also uses at most 15 seconds of each clip (1 to 15, default 5), so a clip can be shortened in the join. If you then lay the original voiceover over the montage, every cut after the first shortened clip lands early, and the voice drifts further with each one.
Fix: build the sound on the Video Editor timeline, which has sound layers, and place the voiceover against the final cut, not against the original clips. Keep talking and lip sync clips out of Montage altogether, since their dialogue would be dropped. More in why does my AI video have no sound.
2. The captions were generated from the script
"Generate captions" in the Video Editor turns a script you paste into a caption layer, timed at an estimated 160 words per minute. It does not listen to the audio. A fast speaker runs ahead of the captions, a slow one behind, and pauses are not reflected at all.
Fix: treat generated captions as a first pass. Play the video, find where the drift starts, and correct the timing from there, or time captions in an editor that transcribes speech.
3. The model cut or padded your track
When you give a model your own audio, its length rules decide what is kept:
| Model on 8frame | What happens to your track |
|---|---|
| Creatify Boreal | Only the first part, as long as the clip length you set, is used; a shorter track is padded with silence. The default length is 5 seconds |
| Pruna p-video | The clip follows the track, up to 20 seconds |
| Pruna Avatar (lip sync) | The clip follows the audio, rounded up to the second, up to 35 seconds; longer audio is refused before you are charged |
| Hailuo H3 lip sync | The audio is clipped at about 15 seconds |
| Clips, music video | Audio longer than 15 seconds is trimmed |
A 12-second voiceover on Boreal at its default length keeps only its first 5 seconds. Lay the full track over that clip later and nothing after second 5 matches.
Fix: set the clip length to the track before generating where the model allows it, and split longer reads into segments that fit. Prices for lip sync by length are in AI lip sync price by model, and the models that take your own audio are in which AI video models can use your own audio.
4. Your editor and your footage disagree on frame rate
Every AI video file we measured on 8frame is H.264 in MP4 at 24 frames per second, and 8frame does not re-encode it: the download is the model's own file. Your other footage may not be: check the frame rate of your phone recordings and screen captures, and whether it is constant. When clips with different rates share one timeline, the editor has to convert some of them. If the sound only starts to drift after you mix footage, look there first.
Fix: set your project to the frame rate of most of your footage, and convert odd clips to a constant frame rate before importing. HandBrake's documentation describes constant frame rate as "not recommended except in special circumstances", and gives import into an editing program as one of those circumstances.
5. The dialogue was generated slightly off
Inside a single clip with generated speech, the timing of the mouth against the voice is the model's work. 8frame has no tool that re-syncs the mouth on footage that already exists.
Fix: regenerate with a shorter line; one sentence per clip leaves less room for drift. If you need exact sync to a recorded voice, use a lip sync model, which builds a new clip from a still portrait and your audio.
FAQ
Why does my voiceover drift after I join AI clips? Montage drops each clip's sound and can shorten clips, so a voiceover timed to the original clips no longer matches the cuts. Re-time it against the final edit on the Video Editor timeline.
Why are my auto-generated captions out of sync? The Video Editor times captions from your script at about 160 words per minute; it does not transcribe the audio. Adjust them by ear.
Why was the end of my audio cut off? On Creatify Boreal the clip length decides how much of your audio is used, and it starts at 5 seconds. Hailuo H3 lip sync stops at about 15 seconds.
What frame rate are AI videos? Every model we measured on 8frame returns 24 fps H.264 MP4.
Can 8frame fix lip sync on an existing video? No. Lip sync tools on 8frame make a new clip from a still image and an audio track.
Sources
- 8frame prices, defaults and limits: the 8frame app, checked 2026-10-21
- Frame rate and codec of AI video files: 8frame's own measurements, video files generated in the 120 days to 2026-09-30, read 2026-09-30
- Constant frame rate for import into an editor: handbrake.fr/docs/en/latest/technical/frame-rates.html, checked 2026-10-21
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.