Score the cut with generated music

Score the cut by asking the Editor Agent to run Score it with music. That automation generates a music bed, lays it under the picture, and ducks it when someone is talking. It spends credits, so you see a plan and you OK it before anything is made. It is not a file picker, not a loudness plugin, and not generate a song from a prompt. Give it a locked cut that already has a voice, a mood in plain words, and a confirmation on the plan.

The same mix also lives on add music to a video and on add music or a voiceover to a video. Those are other doors. The chip in the editor generates, places, and ducks in one pass. Pick one door per session. Do not generate a song, score the cut, then drop the same file from the Audio tab.

What the automation actually does

Score it with music sits in the Audio & music group of the Editor Agent: voiceover, music beds, levels, effects. One click, or one sentence, does three things:

Writes a new bed to the length of the cut
Lays it under the existing audio
Ducks that bed when speech is present

That is all it does. It will not remix stems, match a reference MP3 you have not put on the timeline, or replace the voice. If the picture still has dead air or a bumper you have not committed to, trim the video first. A bed generated against a cut you then shorten no longer matches the picture.

It spends credits. Free timeline edits — trim, split, reorder, a volume nudge — land instantly. Anything that spends credits is shown as a plan and waits for your OK. The plan is the bill, not a listen. You hear the bed after you confirm, on the timeline, under the voice that was already there.

Every plan gets the same models and editor. Scoring is not a Pro gate. You spend credits, not a feature flag.

Run it from the video editor

Open the video editor on the same account you use in the Android app. On a wide screen the Editor Agent opens by default beside the timeline.

A first-day path:

Put a cut on the timeline. Upload, or pull clips from the Video tab. Library items show as strips.
Confirm there is speech. Play it, or click What's on my timeline? and read the rundown. A silent product plate can still take a bed; ducking will have nothing to duck against.
If there is no voice and you need one, write and generate a voiceover first. Score second. A bed with nothing to duck against sits at full level the whole way through.
Click Score it with music, or type the mood in the agent box: "Score it with music: sparse piano, instrumental, no drums, ducked under the voice."
Read the plan. OK it. Wait for the bed.
Listen on a phone speaker, not just headphones.

You can skip the chip and type any instruction in plain words. The chip is a saved sentence. The box is the same agent.

Do not score a cut you still intend to rearrange. Split, trim, and reorder are free and instant. Generate the bed after the picture is locked. If you recut after the bed lands, ducking points sit on the wrong words.

Brief the mood in plain words

The generator only knows what you say. "Make it better" is not a brief. Name the bed the way you would name it to a colleague who cannot see the cut.

Useful briefs:

Score it with music, warm lo-fi, even energy, no drop, instrumental
A quiet acoustic guitar bed under this talking-head, duck whenever I speak
Score it with music: sparse piano, no drums, no vocal samples

Pack four things into that sentence:

Texture. Piano, lo-fi drums, synth pulse, acoustic guitar. One texture. Two is how the bed changes personality halfway through.
Energy. Even, or a lift into the last third. "No drop" is a real instruction. A chorus on your CTA will bury the CTA.
Vocals. Say instrumental. A sung bed against a talking-head is two people talking. The duck will not save that.
Duck. Say it. "Ducked under the voice" repeats what the automation already does, and it stops a bed written as a hero track.

Do not put a credit number in the prompt — the plan quotes the spend. Do not name a playlist track. This writes a new bed; it does not license someone else's.

If the first bed is the wrong genre, do not ask the agent to "make the drums quieter" and expect a remix of the same file. That is a new Score it with music run, a new plan, a new spend. If the genre is right and only the level is wrong, that is Balance the audio, and that lands instantly.

Ducking, talking-heads, and Balance the audio

Ducking is the reason this automation exists for speech. A bed that stays at one level will win every time a phone speaker folds the mix to one channel. The voice is the product. The bed is furniture.

On a talking-head, ducking should drop the bed when you speak, rise in the gaps and B-roll, and stay out of the same fight as the voice — which is why "no drums" and "instrumental" belong in the brief. If you skip ducking and lay a loud bed, the first three seconds of speech are what the viewer loses. That is also where the hook lives. A bed that wins the open loses the cut.

Balance the audio is the neighbour chip: voice up front, music under it. It does not write a new bed. It does not spend credits. Use it when the song is right and the mix is not.

Pair them in that order:

Score it with music — generate, place, duck. Plan, then OK.
Listen. If the voice is still under the bed, click Balance the audio. Instant.
Listen again on a phone speaker.

A bed that still fights after a balance pass is usually the wrong bed, not the wrong fader:

Lyrics against speech. Re-score, instrumental only.
A drop on the line you care about. Re-score, even energy, no drop.
Full-band drums under a quiet close-mic. Re-score, sparse.
A previous episode's track dragged in from the Audio tab, never ducked. Do not re-score if the series needs that same track. Run Balance the audio on what you already have.

What's on my timeline? is the diagnostic when you cannot hear why it is busy. If the rundown says two music beds, you scored a cut that already had a track. Remove the extra strip in the Audio tab, then balance. Two beds is not more production. It is two people humming different songs.

After the mix is right, caption if the cut speaks. Caption the whole cut in the agent is a separate spend. Audio first or captions first does not matter as much as locking picture first. Do not recut after either pass.

Your own track, from the Audio tab

Score it with music writes a new bed. The Audio tab is how you use a bed you already have.

Open the Audio tab. Library items — uploads, generations, music — show as strips. Drop the one you want. That is a free timeline edit. No plan, no generation. You are placing a file.

Use the Audio tab when:

You already generated a song from a prompt and you like that file
You have a bed from last week's episode and the series should sound like one show
You uploaded a track you are allowed to use
You need the same twenty seconds under three cuts, not three new beds

Then, if the voice is buried, Balance the audio. You still get the voice in front without paying to write a new song.

Do not use Score it with music as a second opinion on a track you already chose. It will not "do that, but ducked." It will write something else and bill you for it. If you want a song that will live across a series, generate once, save the strip, reuse it. If you want a bed that exists only to serve this talking-head, score the cut.

When the mix is done, check the video before you publish it. Mute test the captions if you burned them. Unmute test the bed on a phone. If you cannot hear the first sentence, the bed is still winning.

Three moves on a first cut

Open the video editor. Lock picture. Confirm a voice is on the timeline — play it, or ask What's on my timeline?
Click Score it with music and put the mood in the same sentence: instrumental, texture, even energy, ducked under the voice. Read the plan. OK it.
Listen on a phone speaker. If the voice is still under the bed, click Balance the audio. If the song is the problem, score again with a tighter brief. Do not stack a second generated bed on the first.

That is a finished soundtrack for a talking-head. Export when the first sentence is still the loudest thing in the room.