AI Lip Sync Video: Match Any Mouth to Any Audio
Lip sync covers two jobs that look similar and are not. The first is re-voicing: you already have footage of a person talking, and you want the mouth to follow a new audio track — a translation, a corrected line, a different take. The second is making a still photo speak, where the whole performance has to be generated from one frame. VdoBloom's avatar page does both, and the form changes with the model you pick, because the input a video-to-video sync needs is not the input a talking photo needs.
For re-voicing, you upload a video and the audio you want it to follow, and choose a quality mode — a faster one or a higher-quality one. For talking photos, you upload an image and an audio file and pick from the avatar models: a long-form model that takes audio up to five minutes, a higher-resolution model for shorter clips, and further options built around audio-driven performance. Audio uploads can be up to 100 MB, so a full voiceover track fits without splitting.
What this tool does
- VdoBloom's lip sync tool takes a source video plus a new audio track and returns the same video with the mouth movements matched to that audio.
- The video-to-video sync offers two quality modes: a faster one and a higher-quality one.
- Talking-photo models work from one uploaded image plus audio, and one of them supports audio up to five minutes long.
- A higher-resolution avatar model handles clips of up to about 60 seconds at 720p or 1080p.
- Audio uploads can be up to 100 MB, and the uploaded media keeps the face from the photo or the person already in the video rather than inventing a new one.
- New accounts start with free signup credits, one-time credit packs never expire, and subscriptions are also available for a monthly allowance.
Video re-voicing versus talking photos
If you have the footage, use it. Re-voicing a real performance keeps the micro-expression, the blinking and the head movement that a human did naturally, and the model only has to reconcile the mouth with the new audio. Generated-from-photo clips have to invent all of that, which is why a ten-second talking photo can look excellent while a two-minute one starts to feel uncanny.
That makes clip length a design decision rather than a technical limit. Short talking-photo clips work for an intro, a product line or a single sentence; anything longer is better served by filming the person, or by using one of the presenter flows that scripts a performance rather than re-animating a still.
Source quality decides the outcome
The single most useful thing you can do before opening the tool is choose a better source. For a photo: front-facing, eyes open, mouth unobstructed, even light, taken at about head height, with the face filling a good part of the frame. For a video: one speaker, steady framing, no hands near the face, and enough light that the lips are clearly defined.
Audio matters just as much. Clean speech with no music bed and no room echo gives the cleanest mouth shapes, because the model is following the voice rather than a mix. If your audio is a finished voiceover with music underneath, the sync will still work, but the timing tracks the whole file — including the parts where nobody is talking.
How it works
1.Choose the job, then the model
Re-voicing existing footage needs a video input and the video-to-video sync model. Making a still photo talk needs an image input and one of the avatar models.
2.Upload the media and the audio
One clear, front-facing shot of the person — video for re-voicing, a photo for a talking image. Audio goes in as a separate file, up to 100 MB, and the mouth follows that track.
3.Set the quality mode
On the video-to-video sync, choose the faster mode while you are testing and the higher-quality mode for the version you keep. Resolutions run 720p to 1080p depending on the model.
4.Watch the whole clip before you use it
Check the mouth against the audio at the start, in the middle and at the end. Sync drift, teeth and fast consonant runs are the details worth watching for.
Frequently asked questions
What is the difference between lip-syncing a video and making a photo talk?
With a video, the performance already exists — the expression, the blinks, the head movement — and only the mouth needs to be re-fitted to a new track, which typically produces the most believable result. With a photo, the whole performance has to be generated from one frame, so the result depends heavily on the quality of that frame and on keeping the clip short.
How long can the audio be?
It depends on the model. One long-form avatar model takes audio of up to five minutes, a higher-resolution model is built for clips of about a minute or less, and the video-to-video sync is best with short, clean takes so drift cannot accumulate. Uploads themselves can be up to 100 MB, so length is usually limited by the model rather than the file.
Can I dub my video into another language and keep the sync?
Yes, and that is one of the two common reasons to use this. The video translator has its own dubbing flow that separates your voice from the background, translates every line to fit its timing, and can add lip sync on top — which is the better route when the source is a full talking video in another language. This page is the direct route when you already have the audio you want.
What kind of source video works best?
One person, facing the camera, mouth clearly visible, reasonably steady camera, and no hands or objects crossing the face. Good light on the face helps more than high resolution does, because the model is reading lip detail rather than pixel count. Clips with fast camera moves or multiple people in frame give it much less to work with.
Will the face in my photo still look like me?
The face comes from the image you upload rather than from a description, which is the entire point of uploading a photo. How close the video reads depends on the source: a sharp, evenly lit, front-facing photo holds up well, while a small or partially hidden face leaves the model guessing and the result drifts.
Do I need to time the audio to the video myself?
No. Upload the audio as it is and the mouth is fitted to it. What you should do is check the first and last few seconds — if the audio has a long silence at the start, the person sits still through it, which is correct but looks odd if you were expecting speech to begin immediately.
Can I use the result commercially?
Under VdoBloom's terms you keep the content you upload or generate and you are responsible for it. That covers putting a synced clip in a course, an ad or a social post. The responsibility part is real here: you need the agreement of the person whose face or voice is in the material, and using someone else's likeness or voice without permission is your liability, not the tool's.
Related tools
AI talking avatar: turn a photo and an audio track into a lip-synced video
Upload a photo and audio for a lip-synced talking video. 7 models, from 5 credits for a short clip up to 5 minutes of audio on Kling Avatar.
AI voice generator: type text, choose a voice, get an MP3
Type text, pick from 4 TTS engines and up to 30 voices, get an MP3 back. ElevenLabs caps at 5,000 characters; 5-6 credits per 1,000 chars.
AI Lyric Video Maker: Every Beat Gets a Shot
Turn a song into a beat-synced video with the lyrics on screen — generate the track or upload your own, lock one character, and cut every shot on the beat.
AI UGC Ad Generator
Build a UGC ad in 3 steps: generate or upload a creator, composite your product into the shot, then render a talking 9:16 ad with Seedance 2.
Ready to try it?
Lip sync my video