Lip Sync AI: Match Mouth Motion to Any Voiceover
Lip sync AI on VEME matches mouth motion to any voiceover, dub or song and turns a still photo into a talking avatar at export resolution. You can also start from a speaking video and refit the mouth to a new track while the rest of the face stays put.
- Photo or video input
- MP3 and WAV audio
- Any language
The mouth follows the audio, the rest stays put
Lip sync AI reads the sounds in your track and works out a mouth shape for each one, then redraws the lips and jaw frame by frame so the timing lands on the words.
Lips, jaw and timing
Mouth shapes are matched to the speech sounds in the audio, so syllables open and close the mouth at the right moment.
Identity and expression
VEME's own help text says the model mainly edits mouth and jaw movement, while other facial features stay consistent with the source.
A clear view of the face
The face, and the mouth in particular, needs to be clearly visible. Side profiles, hands near the mouth or heavy shadow make the sync less accurate.
Make a photo talk, or re-voice a video
Before lip sync AI does anything, the Lip-Sync panel asks who should talk. You pick one of two uploads, and that choice decides which models you can use.
Talking photo AI from one portrait
Upload a portrait photo of the person plus the audio. The photo models animate the mouth and may add some head motion, so a front-facing headshot with the mouth relaxed works best. The audio must run between 2 and 300 seconds.
New audio on an existing clip
Upload a video of a person speaking plus the new audio. The video keeps its own movement and only the mouth is refitted. Keep the clip and the audio close in length, because the app rejects pairs that differ by more than its set limit.
What people make with an AI lip sync video
These four lip sync AI jobs come straight from what the VEME tool page shows. Each starts with one face and one track.
Dub one clip into new languages
Record a tour guide once, then swap in Spanish or Japanese voiceovers. The mouth realigns to each track instead of drifting off the words.
EN / ES / JA
Make a photo sing
Turn a single portrait into a singing avatar for a chorus or a short music clip, with each lyric mapped to a mouth shape.
Song
Brief a brand host
Have a spokesperson for a made-up label like Fernleaf Skincare read this week's announcement from a recorded voiceover, with no studio day.
Voiceover
Let a pet tell the story
The VEME page also shows a pet narrator. Animal faces are less predictable than human ones, so try a short line first.
NarrationGive lip sync AI a track it can read
What the panel accepts
- VEME's help text lists common formats such as MP3 and WAV.
- If you add a format the tool does not support, it may offer to convert it to MP3 first, which can slightly lower the audio quality.
- Speech in any language works, since the mouth is matched to sounds, not to a dictionary.
- You can also pick an earlier file from your History instead of uploading it again.
Tips for a cleaner sync
- Clear speech gives the best result. Trim long silences and keep music low under a voice.
- For songs, a vocal-forward mix reads better than a dense track with backing vocals.
- Need a voice first? Make one in VEME's Create Voice-over or Voice Clone tools, then bring the file here.
Three lip sync models in the menu
The model list below is what the Lip-Sync panel showed on October 20, 2026. The menu can change, so check it in the app before you run.
| Model | Face input | Audio | Extra field |
|---|---|---|---|
| latentsync | Video of a speaking person | Required | None |
| kling-avatar-v2-standard | Portrait photo | Required, 2 to 300 s | Short text description of the avatar video |
| kling-avatar-v2-pro | Portrait photo | Required, 2 to 300 s | Short text description of the avatar video |
Each model shows its own credit cost in the app, and the two avatar models list a higher cost for Pro than for Standard. Open the Lip-Sync panel to compare them on your own files.
Set up the face for lip sync AI
Face checklist
- One person, facing the camera or close to it, with the whole mouth in view.
- Even light on the face. Strong shadow across the lips hides the shapes the model needs.
- No hands, microphones or hair covering the mouth.
- For video, pick a stretch where the head does not turn fully away.
What to expect, plainly
- Generation uses credits. Sign in and check the cost shown in the tool before running it; account offers and plan details are shown on the pricing page.
- Results are generated, so teeth, tongue and fast syllables can look soft. Watch the clip before you publish it.
- Only use faces and voices you have permission to use, and label synthetic speech where your platform asks for it.
Three steps to a synced clip
Choose who talks
Open Lip-Sync and upload either a portrait photo or a video of the person speaking. The model menu updates to the ones that accept that input.
Add the audio
Upload the voiceover, dub or song, or pick it from History. On the photo models, add a short description of the avatar video you want.
Generate and check
Run it, then watch the lip sync AI result on hard sounds like p, b and m. If something slips, clean the audio or try another model and run it again.
Lip sync AI questions
How does lip sync AI work?
Lip sync AI listens to the speech in a track, splits it into sounds and predicts a mouth shape for each one. It then redraws the lips and jaw on your photo or video in time with those sounds. On VEME the rest of the face is meant to stay as it was.
Can I make a photo talk?
Yes. Choose Upload image, add a portrait photo and an audio file between 2 and 300 seconds long, then run one of the Kling avatar models. A clear, front-facing face gives the most natural result.
Is there a free way to try it?
Generation uses credits. Sign in and check the cost shown in the tool before running it; account offers and plan details are shown on the pricing page.
Which audio formats work?
Common formats such as MP3 and WAV. If you upload something the tool does not support, it may ask to convert the file to MP3 first. Clean speech with little background noise gives the best sync.
Does it work in other languages?
Yes. Matching works on the sounds in the audio, so any spoken language can drive the mouth. That is what makes it useful for dubbing one clip into several languages.
Will the rest of the face change?
The model mainly changes mouth and jaw movement. Other features stay consistent with the original. On the photo models, some head and expression motion may be added so the avatar does not look frozen.
Give your face a voice
Upload a photo or a clip, add the audio and get a synced result in one run.
Start syncing