Strip-and-replace audio only after picture is allowed to freeze
attach_audio_to_video in replace mode mutes the original and lays a new track. Doing that on a miss, or before picture lock, destroys a soundtrack you might have needed and bills a swap you will repeat.
Replace mode is a condemnation. attach_audio_to_video with replace mutes the original and puts your URL in its place — music, VO, SFX, a cleaned isolate. There is no partial duck, no fade of the old track. That is the point. It is also why you do not do it first. A miss with native audio still tells you whether the take had a usable room or a usable line. Mute it early and you are scoring a ghost.
Replace the audio in a video is the job. Same tool as adding music, other mode. Flat credit fee, length-agnostic. The replacement file must already be an HTTPS URL. Local paths fail.
You are throwing the original away on purpose
If you wanted to keep any of it, that is mix with a low original_volume. People use replace as a panic button on buzz, then regenerate the picture, then replace again. Two mutes. Two attaches. The first mute taught them nothing they could not have learned by listening.
Picture lock here means the frames will not be sampled again for taste. Audio lock can be later — you may still isolate, dub, or VO after the picture is frozen. Replace is the moment you assert the original track has zero future. Do not assert that about a candidate.
Native-audio models make this tempting. The line is almost right. Replace "fixes" it. Then you hate the eyes and regen. The replacement track is now an orphan, and the new take has new native sound you might have kept.
When replace is the right later step
- Scratch VO on a locked product shot, swapped for the real read.
- A music-only bed on a MOS plate you already signed.
- A cleaned isolate laid back onto the same locked picture.
The AI video generator produced the file. Replace is not a generate. It is a verdict on the soundtrack of a file you are keeping.
If you are still deciding mix versus replace, you are not ready to attach. Listen to the keeper. Then pick a mode.
Do not replace as a preview
You cannot "try" replace without spending the attach. If you need to hear a bed against picture without committing, that is an editorial preview in a tool that does not mute the source, or a local player with two tracks. Versely's replace is the commit.
estimate_cost if you want the number. It will not save you from doing it twice.
FAQ
Can I undo replace and get the native track back?
Not from the output file. The original is in the source you started with. Keep the source. Replace produces a new video. If you deleted the source, the native room is gone.
Why not always mix at original_volume zero?
Because that is replace with extra steps and a chance to set the number wrong. If the original should be silent, say replace. If it should whisper under a VO, say mix.
Does this re-time the picture to the new audio?
No. It swaps the track. If the new VO is longer than the clip, you needed a different picture or a shorter read — decided before attach.
Should I caption the native take, then replace?
Never. Captions follow the soundtrack you will ship. Replace first on the locked picture, then caption.