Lip sync re-animates a mouth to match new audio. Use it for a script change without a reshoot, to finish a dub so it stops reading as dubbed, or to make a still portrait speak.
Get the audio ready first Whatever they should say has to exist as a file: a recording, or a voiceover you generate under Generate → "Voiceover". Settle the performance before you spend anything on the picture.
Open Generate and pick "Make a face talk" Search for "lip sync" if the label is not the one you expected. The same goal handles a video clip and a still portrait.
Attach the face and the voice One clear, front-on face works far better than a crowd or a profile, because the model has to find a mouth before it can move one. Then attach the audio track it should follow.
Generate and check the join Watch the result at full size before you build on it. Mouth work is where a generation is most likely to look almost-right, and it is easier to re-roll than to fix.
Tips
A still portrait plus a generated voiceover is the cheapest talking-head video there is: no camera, no studio.
If the result looks uncanny, try a tighter crop on the face: more mouth, fewer pixels of everything else.
For a translation, dub first and lip-sync to the dubbed track. Dubbing on its own leaves the original mouth movements.