lipsync

Run the runcomfy CLI with the Sync Labs endpoint: runcomfy run sync/sync/lipsync/v2/pro with a JSON input containing video_url and audio_url. The source video provides everything except the mouth, which is synced to your audio track.

Use Sync Labs sync v2 Pro for mouth-swap on existing video, OmniHuman for a talking-head video from a portrait still plus audio, and Kling lipsync text-to-video when you only have a script with no audio file.

Yes, ByteDance OmniHuman (bytedance/omnihuman/api) takes one portrait image and one audio file and generates a video where the subject speaks and gestures naturally. Head-and-shoulders or upper-body framing works best.

Yes, Kling lipsync text-to-video and HappyHorse 1.0 generate speech audio in-pass from your script and sync it to the resulting video. However, you cannot lock the mouth to a specific pre-recorded voiceover this way.

Drift usually comes from a significant audio/video duration mismatch, so trim the audio or extend the video first. Poor audio quality such as a music bed under the voice also degrades mouth sync; isolate the voice stem if needed.

You must have consent from both the person whose face appears and the speaker whose voice is used. Requests targeting real public figures without consent, or aiming at defamatory or sexually explicit synthetic media, should be refused.