What is AI lip sync from a photo?

It takes one still portrait and one audio track, then generates a video in which that face speaks the audio with matching mouth movement. You are not editing existing footage - the motion is generated from the photo.

Can I lip sync or dub an existing video?

Yes, with the Video Lip Sync model. Switch to the Video tab, then upload your video (MP4 or MOV with H.264 recommended, at least 360p, up to 100MB; other codecs are transcoded) and a clean voice track of up to 60 seconds, and the mouth is re-synced to the new audio. The output keeps your video's resolution, with anything above 1080p scaled down to 1080p, and runs as long as the audio: a longer video is trimmed, a shorter one is looped. Choose Front-facing, single speaker for a faster render, or Complex scenes, multiple shots for footage with cuts. MiniMax H3 Max, the default model, works from a still photo only.

What video and audio does Video Lip Sync accept, and what comes back?

Source video: MP4 or MOV, 360p to 1080p, 24 to 60fps, up to 100MB on our side. Anything above 1080p is scaled down to 1080p, and below 360p is not supported. Audio: MP3, WAV, M4A, AAC or OGG, a clean voice track without music, up to 60 seconds and 10MB. The result is an MP4 at 25fps that keeps the source resolution and runs exactly as long as your audio.

What are the advanced settings on Video Lip Sync?

Remove background sound runs vocal separation first, which helps when the track has music or room noise, and it is available in both scenes. Detect shots and speakers belongs to Complex scenes, multiple shots: it is on by default there and splits the footage at its cuts, and you can turn it off for one continuous take. The remaining three only apply to Front-facing, single speaker: Loop the video to fit the audio is on by default and repeats your footage when the audio outlasts it, Loop back and forth plays each repeat in reverse so the jump back is less obvious, and Start the video at skips the opening seconds of your footage. In Complex scenes the looping stays on and there is no start time. You never need a script or subtitles - the mouth follows the recording you upload.

Do I need to bring my own audio?

Yes. The panel takes an audio file or a microphone recording. If you don't have a recording, generate one from text with our Text to Speech tool, download the MP3, and upload it here. Anything you would record for a voiceover works, including singing and non-speech sound.

How long can the clip be?

From a photo, between 5 and 14.8 seconds. Audio shorter than 5 seconds is rejected by the model. Audio longer than 14.8 seconds is not rejected - it is clipped to the first 14.8 seconds, and you are charged for the clipped length, not the whole track. With Video Lip Sync, the clip runs as long as your audio, up to 60 seconds.

What kind of photo works?

A clear portrait with a visible, roughly front-facing face. The one hard constraint is aspect ratio: it has to be between 0.4 and 2.5, so very tall or very wide crops are rejected before you are charged. Square and standard portrait or landscape photos are all fine.

What does the transcription toggle do?

With it on, the model transcribes your audio and uses the words to guide mouth shapes, which generally sharpens spoken dialogue. With it off, it syncs to the audio waveform alone. Off can be the better choice for singing or for audio that is not speech.

How many credits does it cost?

Credits scale with resolution and the length of your audio: 10 per second at 480P, 16 at 768P, 32 at 1080P and 64 at 2K. A 10 second clip at 768P costs 160 credits. Video Lip Sync costs 8 credits per second of audio, rounded up to the whole second, and credits for a failed render are refunded. The panel shows the exact figure before you generate.

AI Lip Sync for Photos and Videos

Make a portrait photo speak with MiniMax H3 Max, or re-sync the mouth in a video you already have with Video Lip Sync. Add an audio track and the mouth follows your recording.

Modelo

Cargando componente del modelo...

Vista previa

Clips made with this exact tool

Both clips below started as a single still photo and one audio track — nothing else. Turn the sound on; the point is whether the mouth matches the voice.

Talking-head ad read

12s

A 768x1024 portrait plus a 12.05-second voice track. The output came back at 768x1024 and 12.05 seconds — the canvas follows the photo, the length follows the audio.

Spokesperson intro

11.5s

Same setup, a different face and an 11.50-second read. Head movement and blinking are generated too, so it does not read as a static photo with a moving mouth.

Generated on MiniMax H3 Max Lip Sync at 768P. The portraits are AI-generated and the voices are synthetic — no real person is depicted. Both clips are compressed for the web; the originals are about 10 MB each.

What these two lip sync models actually do

Make a photo speak, or put a new voice in footage you already shot.

A still photo starts speaking

Upload a single front-facing portrait and MiniMax H3 Max animates the mouth to your audio. No avatar to build, no rig to set up, no video footage needed.

Re-dub a video you already have

Video Lip Sync re-syncs the mouth in real footage to a new voice track. The result keeps your video's resolution up to 1080p and comes back as MP4 at 25fps, as long as your audio: a longer video is trimmed, a shorter one loops.

Photo clips follow your audio

There is no duration to guess at. MiniMax H3 Max returns a clip exactly as long as the audio you gave it, up to 14.8 seconds, so the sync never drifts out at the end. Video Lip Sync follows your audio too, up to 60 seconds.

Two scenes, one switch

On the Video tab, Front-facing, single speaker renders fastest on talking-head footage and unlocks the looping and start-time options. Complex scenes, multiple shots detects the cuts and identifies the speaker in each shot, so wide shots and edits are handled instead of smeared - and you can switch that detection off in Advanced settings when the footage is one continuous take.

Transcript-guided mouth shapes

The model can transcribe your audio and use the words to drive the mouth, which sharpens consonants in spoken lines. Switch it off and it syncs to the waveform instead.

480P through 2K on photos

On the Photo tab, draft at 480P, work at 768P, and render at 1080P or 2K when the clip is going somewhere public. 2K is available on lip sync specifically. Video Lip Sync has no resolution picker: it keeps whatever your footage is, up to 1080p.

How to lip sync a photo or a video

Three steps in either tab, and you bring your own audio.

Pick a tab and upload

Photo takes one clear, front-facing portrait with an aspect ratio between 0.4 and 2.5. Video takes MP4 or MOV footage at 24 to 60fps, 360p and up; anything above 1080p is scaled down to 1080p.

Add your audio track

Upload a recording or capture one from your mic. Photo needs at least 5 seconds and clips anything past 14.8. Video accepts up to 60 seconds and keeps all of it.

Set the options and generate

Photo picks a resolution from 480P to 2K. Video picks the scene mode, and Front-facing, single speaker adds looping and a start time under Advanced settings. Both show the exact credit cost before you spend anything.

AI Lip Sync FAQ

What it does, and what it does not.

Give a photo a voice, or a video a new one

Upload a face photo or the footage you already shot, add your audio, and get a lip-synced clip back.

See how it works