The Master Guide / 13 Sound design and mix · Where sound comes from
Keeping or muting in-shot sound
The shot's own sound stays where it works. Where it fails, mute that part; never pay for a new take to fix the sound (decided).
The shot's own sound stays where it works. Where it fails, mute that part; never pay for a new take to fix the sound (decided).
Keep it when it is the right sound, in sync with the action, with no music bleed and no garbled speech.
Mute the failed part, and replace it from the sound list, when it has:
music or a score nobody asked for (only music whose source is in frame is allowed, such as a radio or a band on stage, kept low under dialogue);
voices, murmur or garbled speech in a shot without dialogue;
the wrong sound for the action, or a sound that fights the line;
a named unwanted event (a phone ringing, footsteps where none belong);
any Arabic voice or crowd murmur (walla), in any shot. Seedance never voices Arabic, walla included (decided). Replace Arabic murmur with a real CC0 crowd recording (see Sound design and mix → Real recordings from Openverse (CC0)) or with voices recorded in ElevenLabs (see Speaking shots and lip sync → Other speaking routes).
Listen to every shot's sound, and read a transcript of it (ElevenLabs Scribe) to catch murmured words you might miss. A video-analysis tool can also report whether the sound fits the action; it reports and never decides. Generated sound can bleed across a cut in a multi-shot clip, so listen at every cut.
Sound-off speaking plates (speaking shots generated without sound, for lip sync; see Speaking shots and lip sync → Route 2, the Arabic default) have no sound of their own. Their room tone, ambience and foley come from a designed effect, a neighbouring shot's own sound or a real CC0 recording from Openverse. The same bed is the room tone under a lip-sync output. This is a starting practice; no source covers it.
Mark the shots whose own sound is used in the mix (on-screen dialogue, lip sync, featured foley). Only those carry their earlier takes' sound beneath them in the timeline (see Editing → The editable timeline).