Contents
To make an AI lip sync video with MiniMax H3 Max Lip Sync, upload one portrait and one voice track, choose a resolution, and generate. The model moves the mouth to match your audio and returns a talking video that runs exactly as long as the recording, at up to 2K.
There is no prompt to write and no duration to set, so the result comes down to two inputs: the face and the voice. This guide covers how to prepare both, including which voice model to use, the step-by-step workflow in Picsart's AI Playground, what each setting does, and how to fix the most common problems.
What MiniMax H3 Max Lip Sync does
MiniMax H3 Max Lip Sync is the talking-head variant of the MiniMax H3 Max family. The other members generate video from a prompt or an image; this one has a single job, turning a still face into a speaking one. It follows your uploaded track frame by frame, so the tone, accent and timing of the voice stay exactly as recorded.
- Input: one portrait, used as the first frame, plus one audio track
- Length: set by the audio, from 5 to 14.8 seconds
- Resolution: 480p, 768p, 1080p or 2K
- Aspect ratio: taken from the uploaded image
- Optional: a transcript to guide difficult speech
Whoever is in the photo is exactly who speaks in the video, which makes it a reliable way to keep one presenter looking identical across many clips.
Why audio-first matters
Many AI video models generate the voice along with the picture, so the performance is whatever the model decides. Here the order is reversed: you settle the voice first, then the face is animated to match it. That gives you full control over the exact words, the delivery and the language before a single frame is made. It also means a weak recording cannot be rescued by the video, which is why most of this guide is about preparing the inputs well.
What you need before you start
A portrait
The photo becomes the first frame and the face that speaks in every frame after it, so it is worth choosing carefully:
- Framing: head and shoulders work best for a talking clip. Leave a little space above the head and below the chin.
- Angle: the face turned toward the camera, or slightly off it, with the whole mouth visible.
- Expression: a relaxed or gently closed mouth gives the cleanest, most neutral starting point for speech.
- Light: soft, even light on the face, with the eyes and lips in focus.
- Shape: crop the photo to the shape the video needs, because the output takes its aspect ratio from the image. Use 9:16 for Reels, Shorts and TikTok, 1:1 or 4:5 for the feed, and 16:9 for YouTube and web pages.
The portrait can be a photo you took or a character you generated with an AI image model. Generated portraits are a good fit for presenters and mascots, since you can make the exact face, outfit and setting a brand needs.
A voice track
The audio decides everything about the performance. There are three ways to get one:
| Route | Best for | What it gives you |
|---|---|---|
| Your own recording | Founders, creators, real spokespeople | Your real voice, delivery and accent |
| Eleven v4 | Presenter voiceovers and translated versions | A voice from a full catalog, audio tags to direct emotion and pauses, 90+ languages |
| Gemini 3.8 Flash TTS | Characters and mascots | A voice you describe in plain words, accents on request, 130+ languages |
Recording it yourself. Use a quiet room, keep the microphone a hand's width from your mouth, and record the line dry, with no music underneath. Read it two or three times and keep the most natural take.
With Eleven v4. Paste the script, pick a voice from the catalog, and add audio tags where the line needs a pause, a laugh or extra emphasis. The stability and similarity controls keep the same voice consistent across a series of clips, and the same script can be voiced in another language for a translated version.
With Gemini 3.8 Flash TTS. Describe the voice instead of picking one, such as a warm, unhurried narrator in her fifties or an excitable teenage gamer, and the model performs the script in that voice. You can ask for a specific regional accent and direct each line, which makes it a strong match for a character who needs a voice nobody else has.
Whichever route you use, trim the file to between 5 and 14.8 seconds. At a natural speaking pace that is roughly 12 to 35 words: enough for a hook, a product line, a tip or a short greeting.
A transcript, only when needed
Most tracks need nothing else. Keep the written words of the line ready for the takes that are hard to follow: fast delivery, a strong accent or a noisy recording. The transcript guides the sync on exactly those takes, so there is no need to write one for every clip.
Step by step: from portrait to talking video
Open MiniMax H3 Max Lip Sync from the model list in Picsart's AI Playground, then follow these steps.
1. Upload the portrait
Add the photo as the starting frame. Its proportions set the shape of the finished video, so there is no separate aspect ratio to pick. Check that the face is sharp and the mouth fully visible before moving on.
2. Add the audio
Upload the voice track of 5 to 14.8 seconds. The clip will run exactly as long as this file, with no silence added at the end, so trim any dead air at the start and finish before uploading.
3. Choose the resolution
Pick 480p or 768p for quick test takes, and 1080p or 2K for the version you publish. Turn on the transcript only for a difficult recording, such as rapid speech, a heavy accent or background noise.
4. Generate and check the sync
Generate the clip and watch the mouth on b, p and m sounds, where the lips should fully close, and on long vowels, where the mouth should stay open. Keep the same portrait and swap the audio to make the next take, so the presenter looks the same in every version.
The settings, explained
| Setting | Options | When to change it |
|---|---|---|
| Resolution | 480p, 768p, 1080p, 2K | Low for drafts and tests, high for the final cut |
| Aspect ratio | Follows your photo | Crop the photo first to change it |
| Length | 5 to 14.8 seconds, set by the audio | Trim or split the recording |
| Transcript | Off or on | Turn on for fast, accented or noisy speech |
A simple rule for resolution: test at 768p until the sync and the read are right, then render the keeper at 1080p for social or 2K for a large screen, a website hero or a presentation.
How to make a lip sync video longer than 15 seconds
Because the audio sets the length, a script longer than 14.8 seconds needs splitting into takes. Break it where a speaker would naturally pause, at the end of a sentence or a thought, and keep each piece between 5 and 14.8 seconds.
Generate every take from the same portrait so the face matches from cut to cut. Then line the clips up in the editor, and add captions, music or a b-roll shot over the joins if you want them to feel seamless. A 45-second explainer, for example, becomes three or four short takes from one photo.
Five ways to use AI lip sync with MiniMax H3 Max Lip Sync
A recurring virtual presenter
Generate or shoot one portrait, then give it a new script every week for tips, updates or announcements. The face never changes, so a series feels consistent without booking anyone again. A line like "Here's this week's quick tip for better product photos" fits easily into one take.
Talking-head narration
Put a face to a voiceover you already have, such as an explainer, a tutorial intro or a podcast teaser. A speaking presenter gives a short clip a clear point of focus, where a still image with audio underneath can feel unfinished.
One presenter in several languages
Keep the same portrait and swap in a translated recording for each market. Voice the script in Eleven v4 in each language, generate one clip per track, and every version shows the same presenter speaking the local language, with the mouth matched to that track.
Character dialogue
Give a character you generated or designed a voice and a line. Describe the voice in Gemini 3.8 Flash TTS, such as a gravelly old sea captain or a cheerful robot shopkeeper, and the character speaks in it. It suits story teasers, game characters and mascot-style brand faces.
Short social clips
A 5 to 15 second talking clip fits a hook, a reaction or a greeting. Make a vertical portrait, record a punchy line such as "Stop scrolling, you need to see this," and post it as a Reel, a Short or a TikTok.
Tips for cleaner lip sync
Record dry
A clean voice track with no music underneath is the simplest input. Add music later in the editor.
Keep the mouth clear
Avoid photos with a hand, hair, microphone or mask in front of the lips.
Write lines to be spoken
Short sentences with a natural pause read better than long, dense ones.
Check translated timing
The same line can run longer in another language, so trim it to fit under 14.8 seconds.
Test low, publish high
Try the take at 768p, then render the keeper in 1080p or 2K.
Troubleshooting common lip sync problems
The mouth drifts on fast lines
Rapid or mumbled speech gives the model less to follow. Turn on the transcript for that take, or re-record the line a little slower with clearer consonants.
The clip is shorter or longer than expected
The video always runs as long as the audio file, including any silence at the start or the end. Trim the dead air before uploading and the clip will match the spoken line.
The video is the wrong shape
The output takes its proportions from the photo. Crop the portrait to the ratio you need, such as 9:16 for Reels or 16:9 for YouTube, and generate again.
The voice sounds flat
The video can only be as lively as the recording. Re-record with more energy, or add audio tags for emphasis, pauses and laughs in Eleven v4.
The face looks different between takes
Each take starts from whatever portrait you upload. Reuse the exact same image file for every clip in a series so the presenter stays identical.
The line does not fit
A clip tops out at 14.8 seconds of audio. Split a longer script into takes at natural pauses and join them in the editor.
Where it sits in the MiniMax H3 family
Lip Sync is one of six MiniMax H3 models in Picsart. MiniMax H3, MiniMax H3 Max, MiniMax H3 Max Turbo, MiniMax H3 Max Extend and MiniMax H3 Max Camera Controls each handle a different step of a video:
| Model | What it does |
|---|---|
| MiniMax H3 | Native 2K video with synchronized audio from a prompt or an image |
| MiniMax H3 Max | Fast text-to-video and image-to-video with start and end frames, 480p or 768p |
| MiniMax H3 Max Turbo | The fastest tier, for rapid iteration |
| MiniMax H3 Max Extend | Continues an existing clip by 5 to 15 seconds |
| MiniMax H3 Max Camera Controls | Turns one image into a programmed camera move |
| MiniMax H3 Max Lip Sync | Turns a portrait and a voice track into a talking video |
One way to combine them: build the presenter's scene with MiniMax H3, then use Lip Sync for the close-up where they speak to camera.
Frequently asked questions
It is the lip sync variant of MiniMax H3 Max. It takes one portrait and one audio track and returns a talking video with the mouth matched to the sound, at up to 2K.
Make your first talking video
Pick a portrait, record or generate a line, and open MiniMax H3 Max Lip Sync in the AI Playground to make your first talking video.
