Overview

Sync-3 is a lip sync model built for full-scene generation. Instead of stitching clips together, it processes the entire shot in one go. That means more consistent results and fewer visual glitches.

It keeps the speaker’s style and emotion intact across 95+ languages. It also handles tricky situations like close-ups, extreme angles, and partially hidden faces or mouths.

How it works

Understanding the input

Sync-3 works with video and audio, not text prompts. It looks at the whole sequence to understand the speaker across every frame. This helps it keep expression and movement consistent from start to finish.

Because it reads the full scene, it knows what to change and what to leave alone.

Lip sync

The model matches mouth movements to your audio. It processes the full shot at once, so the result feels smooth and natural without awkward transitions.

Scene awareness

Sync-3 sees more of the scene than earlier models. That wider view helps it handle angles and obstructions around the face while keeping movement believable. You’ll spend less time fixing issues and rerunning outputs.

Key features

• Full-scene lip sync

  Processes the entire shot in one pass for consistent results.

• Global face understanding

  Keeps identity, expression, and movement stable across frames.

• Obstruction handling

  Works well even when parts of the face or mouth are hidden.

• Multi-language support

  Maintains style and expression across 95+ languages.

• Wider scene context

  Uses more visual context to improve accuracy.

How to use

1. Upload your video (MP4) and the audio you want to sync (WAV or MP3).
2. Choose the model and endpoint.
3. Run the sync.
4. Download your finished video.

For best results, use clear audio and high-quality video with a visible face.

Documentation

Full docs, parameters, and examples: https://runware.ai/docs/models/sync-3