MiniMax H3 Max Lip Sync Video Generator
Choose a portrait, add dialogue, speech or a recording, and create a 5–15 second H3 Max video.
What MiniMax H3 Max lip sync creates from a photo
H3 Max turns one portrait into a new talking video. The photo guides appearance; your text sets the dialogue and scene. Optional reference audio guides the voice and timing. The result is an AI-generated performance, not a recording of the person.
Use this reference-to-video workflow for a brief introduction, product message, or character scene. Keep each clip focused on one speaker and one idea. To change mouth movements in an existing video, use the separate video workflow.
Text or reference audio: choose how to direct the voice
Choose text for a new scripted line, or reference audio for an existing spoken performance. Both modes require a prompt describing the scene and delivery.
| Decision | Text generation | With reference audio |
|---|---|---|
| What you provide | One portrait plus dialogue and scene text. | The same portrait and text, plus one 2–15 second audio clip. |
| What sets the duration | Choose any whole second from 5 to 15, in 1-second steps. | The detected audio length, rounded up to a whole second, with a 5-second minimum. |
| When it is useful | Trying a new line, greeting, or short scripted message. | Working from an existing voiceover or a performance you have permission to use. |
Reference audio guides the performance; it does not guarantee an exact voice clone. Check names, pronunciation, and delivery before publishing.
How to make your first H3 Max talking video
01
Choose the portrait and optional audio
Upload one portrait. For reference audio, trim a complete spoken segment to 2–15 seconds before uploading; this page does not include an audio editor.
02
Write the dialogue and scene direction
Write the spoken line, expression, and camera direction. Keep the text consistent with any reference audio, then confirm permission to use your materials.
03
Review the exact credit charge
Choose the resolution and text-mode duration. Uploaded or recorded audio is automatically checked after sign-in. Wait for the exact charge, then select Generate. Text-to-speech shows the maximum credit hold.
04
Generate, review, and download
Submit once and follow the job status. When complete, download and watch the video with sound before sharing. Completed results are also available in Activity.
Prepare a portrait and audio that are easy to work with
Portrait checklist
- Choose one person, facing forward or at a gentle angle, with the face large enough to see clearly.
- Keep lips free of hands, microphones, hair, and heavy shadows.
- Prefer a clean image over heavy sharpening or beauty filters. Generated facial details can still vary.
Reference-audio checklist
- Use one speaker and a complete phrase; trim long silence without cutting off words.
- Avoid overlapping voices, loud music, and heavy background noise. Listen through the recording before uploading.
- Check that the clip is 2–15 seconds. A shorter valid clip still uses the 5-second minimum output.
See the lip sync quality guide for more framing and source-quality advice.
Write an H3 Max prompt with dialogue, action, and camera direction
Write what the person says, how they perform it, and what the camera shows. Start with one spoken line, one expression or gesture, and a simple shot:
The person looks into the camera and says, “Welcome. Here is one idea you can try today.” Friendly, calm delivery with a small smile. Keep the camera steady and the face clearly visible.
Quotation marks separate speech from direction. “Friendly, calm delivery” sets the performance; “keep the camera steady” sets the shot. With reference audio, follow its dialogue rather than introducing a different line.
Read the line aloud. If it feels rushed in five seconds, shorten it or choose a longer text generation. Avoid combining several scene changes and gestures. Change one instruction at a time when comparing results.
H3 Max credit examples: know the charge before you generate
Billing follows generated duration, not prompt length. 480P costs 4 credits per second; 768P costs 7:
| Input | Billable video duration | 480P credits | 768P credits |
|---|---|---|---|
| Text · 5 seconds selected | 5 seconds | 20 | 35 |
| Text · 10 seconds selected | 10 seconds | 40 | 70 |
| Text · 15 seconds selected | 15 seconds | 60 | 105 |
| Audio · 2.4 seconds detected | 5 seconds | 20 | 35 |
| Audio · 7.2 seconds detected | 8 seconds | 32 | 56 |
Audio rounds up to a whole second, with a 5-second minimum. Clips outside 2–15 seconds are rejected, not silently trimmed. Confirm the exact charge after detection; file names are not a reliable duration estimate. Text-to-speech holds up to 60 credits at 480P or 105 at 768P and settles after speech is measured; unused credits are returned, and failed jobs release the entire hold.
Compare credit plansReview your result and make focused improvements
Watch the full clip with sound. Check the words, mouth movement, facial details, and camera. A completed job does not guarantee that every creative instruction was followed exactly.
The dialogue feels rushed
Shorten the sentence or extend the text-mode duration. For audio, use a naturally paced recording with matching written direction.
The face or mouth is difficult to read
Recheck the portrait against the checklist above, then simplify head movement and camera direction.
The scene does too much at once
Reduce the prompt to one action and one shot. Remove extra turns, product reveals, or cuts.
A generation fails or is still processing
Check the existing status before resubmitting. Failed jobs return reserved credits; regenerating a completed video is a separate charge.
For job-status or credit-return issues, share the generation details with support. Get help with a generation
Use authorized media and tell viewers when a realistic performance is AI-generated. Our privacy policy explains how uploads are handled. Read the privacy policy
Prompt examples
Start with dialogue, then direct the performance
Product explainer
The person looks into the camera and says, “This feature turns one portrait into a naturally speaking video.” Confident delivery, steady camera.
One sentence covers one feature. Delivery and framing instructions support the message without adding competing selling points.
Story character
The person turns slightly, smiles, and says, “I have been waiting for you to come back.” Warm interior light, restrained facial movement.
The line carries the emotion; small movements support it. Check character continuity before combining independently generated scenes.
Social clip
The person says excitedly, “Are you ready? Let’s begin.” Quick pacing, a small hand gesture, facing the camera.
A short opening leaves room for expression. Check framing and cropping for your chosen social platform before sharing.
FAQ
Before you generate with H3 Max
Understand the media scope, exact credit charge, audio limits, and automatic failure refunds.
Create your H3 Max video from one portrait
Review the charge or speech credit hold before submitting a private generation job.
Visemix is an independent service and is not affiliated with or endorsed by MiniMax or fal.