Media Lipsync (open-source talking-head + lip-sync)

Context: $ARGUMENTS

Quick start

• Animate a photo from audio (head pose + expressions + mouth): scripts/lipsync.py animate --model liveportrait --source face.jpg --driving speech.wav --out talking.mp4 → Step 3.
• Animate a photo from a driving video (retarget expressions): scripts/lipsync.py animate --model liveportrait --source face.jpg --driving actor.mp4 --out retarget.mp4 → Step 3.
• Re-sync an existing video's lips to new audio: scripts/lipsync.py sync --model latentsync --video original.mp4 --audio new.wav --out resynced.mp4 → Step 4.
• Dub a foreign-language clip with lip correction: scripts/lipsync.py dub --video en.mp4 --new-audio es.wav --out es_dubbed.mp4 → Step 5.
• Install a model: scripts/lipsync.py install liveportrait → Step 2.

When to use

• Drive a still portrait photo with an audio track to produce a talking-head video (full facial motion, not just the mouth).
• Re-sync the lips of an existing video to a new audio track (dubbing, ADR, language swap) without regenerating the whole face.
• Need commercial-safe, open-source weights. Do NOT use Wav2Lip, SadTalker, HeyGen, D-ID, Synthesia, Runway Act-One — all excluded per references/LICENSES.md.

Which model does what (read this first)