Kling Lip Sync

The cheapest model in the clip catalog, and the most narrowly scoped. It takes a video and an audio track and re-times the speaker's mouth to the new audio. It does not generate a shot, animate a still, or change anything but the sync.

That narrowness is what makes it useful. Redubbing a line, fixing a take where the delivery was wrong, putting a clip into another language with a voice model: all of these are jobs where generating new footage would be absurd, and this does them for six credits per five seconds at 1080p.

The video input is required; the audio drives the length, two to ten seconds. It is the last step in a pipeline rather than a starting point: generate or shoot the footage, produce the line with one of the voice models, then bring both here. If the source is a still image rather than footage, this is not the model. P-Video Avatar and OmniHuman 1.5 are the ones that animate a photograph.

Describe what you want in your own words and Kuro writes the prompt for you. If you would rather write it yourself, this is what Kling Lip Sync responds to.

Kling Lip Sync re-syncs the mouth in a video that already exists to new speech. Pick a take where the speaker's face stays visible and unobstructed throughout.

The speech comes from one of two places: a voice track you attach, or — with no track — the prompt itself, spoken as the script. Write it as dialogue, never as a shot description.

Without a track, the script is read in the model's default voice. For a chosen or saved voice, make the voice-over first and attach it.

Cut the speech to fit the source: the clip is as long as the video, not as the words.

What it can't do

• It needs a source video — an MP4 or MOV under 100 MB, between 720p and 1080p, within the allowed length.
• No image input. One source video and at most one audio track.
• No visual direction: the picture, framing and resolution are the source's. Sound is always on.
• Tagging a saved Subject adds only its written description, and without a voice track that description is what gets spoken.

Use something else when…

• There is no footage yet, only a portrait and a voice: OmniHuman 1.5.
• A long recording should be dubbed over a looped performance: Sync Lipsync 2.
• The mouth already says the right words and only the voice should change: use Kuro's change-voice tool on the clip.