Docs menu
On this page

Lipsync

The Lipsync node syncs mouth movement to an audio track. Use it to make a video's speaker match new dialogue, or to turn a still portrait into a talking video from an audio clip.

What it does

Lipsync works in one of two modes, depending on the model you pick:

  • Video driven: you give it a video and an audio track, and it matches the mouth movement in the video to the audio.
  • Image driven: you give it a still image (like a portrait) and an audio track, and it animates the image into a talking video.

Inputs and outputs

  • Image or Video: the source. The label switches between "Image" and "Video" depending on the model you choose, and so does what you can wire in: an image-driven model only accepts image sources, a video-driven model only accepts video sources. Switching the model to the other kind removes a source connection that no longer fits.
  • Audio: every Lipsync model needs an audio track to sync to.

The source and Audio each take one connection. Once one is connected it won't take a second, and hovering it tells you to disconnect it first. A workflow that already has two on one of them won't start until you disconnect one.

  • Prompt: a text box for describing the motion. Only shown for models that support it.
  • Flow: an optional input that controls when this node runs. It can only be fed by the Trigger node or a logic node like Sequence or Router.
  • Video (output): the finished, synced video.

If no video or image reaches the node, the run stops with a message asking you to connect a source. The same happens if audio is missing. A source of the wrong kind uploaded straight into the node (rather than wired in) is caught when the run starts, with a message telling you which kind that model needs.

Settings

  • Model: pick which Lipsync model to use (see below).
  • Resolution: shown only for models that offer a quality tier.
  • Model Settings: opens a panel with whichever settings apply to your chosen model, such as Sync Mode and Seed. A note at the bottom reminds you that framing and aspect ratio always follow your input source, these settings only tune the model itself.
  • Prompt: a text box at the bottom of the card, shown only for models that support one. If a Prompt input is connected, this box is disabled and shows "Connected from input" instead. You can connect more than one LLM, Text or Agent node; their texts are joined in the order you connected them.

Once a result exists, the Download button on the card becomes active.

Click Save to Media, next to Download, to keep a copy in your Media Library, in the folder named after your workflow. It saves the video the card is showing, and the icon turns into a check once it's saved. If your storage is full, the card tells you and nothing is saved.

Models

Lipsync runs on Atlas Cloud, like every node.

  • Sync Lipsync 3 (video driven): Sync Mode options (Cut off, Loop, Bounce, Silence, Remap).
  • VEED Lipsync (video driven): no extra settings.
  • OmniHuman 1.5 (image driven): optional prompt, Resolution 720p or 1080p, and Seed. This is the default model.
  • Kling AI Avatar 2.6 Pro and Kling AI Avatar 2.6 Standard (image driven): optional prompt.
  • InfiniteTalk (image driven): optional prompt, Resolution 480p or 720p, and Seed.

A video driven model stops if you connect an image, and an image driven model stops if you connect a video.

Tips

  • The preview on the card automatically resizes to match your finished video's real shape once it exists.
  • Switching models resets fields like resolution and sync mode to the new model's own defaults. Seed is not reset, it carries over from whatever you last set.
PreviousVoice GenerationNextCompose