Can I make the voice, the music and the effect in one place?

Yes. The audio composer has a Voiceover, a Music and a Sound effect task, each with its own models, and every result is an audio asset in your Library.

Speech is billed by the characters in your script, with a small minimum; the button shows the price as you type. Music and sound effects are one flat charge per request, shown on the button before you generate.

Does the narration line up with the picture automatically?

No. You place it. In Studio each audio asset is its own item on an audio lane, so you slide the take to the frame you want and trim the music to the clip.

Should I just let the video model make the sound?

Some models can, and it is quick. You lose the ability to change one sound without rerolling the picture. Render silent when you want to control the mix.

Does exporting from Studio cost credits?

No. The MP4 render is not billed; only the generations that made the clip and the sounds were.

Can one prompt make two sound effects?

It returns one blend. Make each cue its own generation, then place them separately.

Guides / Audio

Voice, music and sound effects for your video

A clip without sound is half a clip. The audio composer makes the three things a soundtrack needs, one at a time, and Studio puts them under the picture.

10 min read
September 24, 2026

What you make, in order. Tap one to jump to its step.

You make a voice, a music bed and a sound effect as three separate generations in the audio composer, each on the model built for it, and you put them under your clip in Studio. The composer's Voiceover task turns a script into speech, Music writes a track from a description, and Sound effect makes a one shot cue or an ambience. None of them needs the video; they are audio assets in your Library until you lay them on a timeline.

This guide follows one silent clip, a fragrance ad made with Seedance 1.5 Pro with its audio switched off, from nothing to a mixed soundtrack. Every clip you can play here is real output, and the prices come from the live catalog when the page loads.

What goes wrong, and the fix

Mistake	Why it hurts	Fix
Asking a video model for the whole soundtrack	A model with native sound writes its own music and effects, and you cannot change one without rerolling the picture	Render the clip silent, then make each sound on the audio composer and mix in Studio
One long music prompt with a voice in it	Music models sing what you describe; a narration line comes back as lyrics	Keep the script in Voiceover and the mood in Music
A full sentence for a sound effect	The effect model reads a scene and makes a wash of noise	Describe the one sound, its material and its room, in a short line
Choosing the voice by name only	Two named voices on the same model read the same line at very different paces	Play the free preview before you spend, and use the speed field on the API if the take runs long
Re-recording narration to fit a cut	Every rerun costs the characters again	Trim and slide the take in Studio; the clip can hold or extend instead

Before you start

A clip in your Library. For this guide it is five seconds of a perfume bottle in haze, made with Seedance 1.5 Pro with its audio switched off, so nothing fights the soundtrack.
Your script, written out. Speech bills per character, so cut it before you generate, not after.
One line each for the music and the effect. The models take a description, not a file.
Any plan. Every audio model is available on every plan, and speech starts at a two credit minimum on the recommended model.

A faceted black perfume bottle with a gold cap on a wet black floor, a magenta beam from the left and a blue beam from the right cutting through haze, dust drifting — **The silent source.** Made with Seedance 1.5 Pro, audio off, five seconds. This is the clip every sound below was made for.

Make the narration

Open the audio composer. Under What do you need? the Voiceover task is selected by default, with ElevenLabs v3 as the recommended model and a Voice list of ten named voices. Type the line as you want it read and pick a voice; we used Roger. The button reads Generate audio with the live price, and the price follows your script as you type. One click generates.

The script, exactly as sent

Some nights do not ask for permission.

The audio composer with Voiceover selected, the script typed in the prompt box, ElevenLabs v3 as the model, the Voice select on Roger, mp3 as the format, and the Generate audio button showing the live price — **Voiceover on ElevenLabs v3.** Pick a voice from the list; the button shows what this exact script costs.

Tip:Preview a voice for free. Each named voice on ElevenLabs and Kokoro has a short recorded sample. Play it on the Characters page voice dialog before you spend on a take.

Make the music bed

Switch the task to Music. Lyria 3.5 is the recommended model, and the price is one flat charge per track, whatever you type. Describe the mood, the instruments, the tempo and how the track should move; say no vocals if you do not want singing. You get a track longer than the clip, which is fine: you trim it in Studio.

The music prompt, exactly as sent

Dark minimal electronic pulse for a fragrance commercial: a deep sub bass heartbeat, sparse glassy synth stabs, a slow rising pad, 90 bpm, cinematic and expensive, no vocals, no drums until the last two seconds.

The audio composer on the Music task with the track description typed, Lyria 3.5 as the model and the Generate audio button showing the flat price — **Music on Lyria 3.5.** One flat charge whatever you type. Ours came back two minutes long.

Make the sound effect

Switch to Sound effect. ElevenLabs Sound Effects v2 is the recommended model, one flat charge per request, and it takes a short prompt of at most 450 characters. Name the sound, the material and the space it happens in. One sound per generation: a second effect is a second run.

The effect prompt, exactly as sent

A single deep glass chime with a low reverberant thud underneath, decaying slowly in a large empty hall.

The audio composer on the Sound effect task with the chime described, ElevenLabs SFX as the model and the Generate audio button showing the flat price — **Sound effect on ElevenLabs Sound Effects v2.** Ours came back one second long, which is what a chime needs.

Watch for this:MMAudio v2 works from the picture, not from words alone. The second model in the Sound effect list, MMAudio v2, is a video to audio model: it watches a clip and writes a matching track. It is the pick when you want the room and the movement scored automatically, not a specific cue.

Put it under the clip in Studio

Open Timelines, press New timeline, pick a project or name a new one, name the timeline and press Create. Studio opens on it. In the left panel, switch from Project media to Library: that is your whole library, with a search box and type filters. Each card has a + that adds it at the playhead.

Search for the clip and press its +. It lands on the first video lane at 0.
Click the music card so it loads in the Source preview above the list. Drag the preview scrubber to 0 and press Mark In, then to 5 seconds and press Mark Out, then press +. The bed lands on an audio lane trimmed to the clip; the rest of the two minute track never reaches the timeline.
Drag the timeline scrubber to 1.6 seconds and press + on the narration; it lands on its own audio lane at the playhead.
Drag to 3.9 seconds and press + on the chime.

Studio with the fragrance clip on lane V1 and three audio lanes under it: the trimmed music bed on A1 from the first frame, the narration on A2 from 1.6 seconds, the chime on A3 near the end, the Library rail on the left and Export video top right — **One video lane, three audio lanes.** Every sound stays its own item, so you can move or mute one without touching the others.

Press Export video at the top. The MP4 renders on the server, lands in your Library tagged as a render, and is listed under the project's recent exports. The export is not billed.

Tip:The edits line does not block the export. After you add items, Studio shows a line saying edits are waiting for the NOLGIA Agent to apply. That is about writing them into the timeline's authored document. The export already includes them: our render carried all three sounds.

The silent fragrance clip — The same five seconds, exported from Studio. Play the right one to hear the bed, the line and the chime together.

The same fragrance clip with the narration, the music bed and the chime mixed under it — The same five seconds, exported from Studio. Play the right one to hear the bed, the line and the chime together.

Which model for which job

You need	Task	Start with	Also in the list
A narrator or a line of dialogue	Voiceover	ElevenLabs v3, ten named voices	v3 Conversational at half the rate, Multilingual v2, Turbo v2.5, Kokoro with twenty voices, MiniMax Speech, Inworld, Dia, Orpheus
A music bed with no words	Music	Lyria 3.5	MiniMax Music 2.6, ElevenLabs Music, Stable Audio 3 Medium and 2.5
One cue: a hit, a whoosh, a door	Sound effect	ElevenLabs Sound Effects v2	MMAudio v2 when the track should follow a clip
Sound that follows the picture	Sound effect	MMAudio v2	Or a video model with native audio, at the cost of control

The lists as the composer shows them today. The recommended model is marked in the picker.

Where you still regenerate

Levels. Studio plays every lane at full level; there is no volume, fade or ducking yet. If the bed covers the line, move its Mark In to a quieter passage of the track, or ask the music model for a hushed stretch.
A voice you want to keep. Named voices exist on ElevenLabs and Kokoro only; MiniMax Speech, Inworld, Dia and Orpheus publish no voice list here, so a rerun on those may not sound the same. The same voice in every video shows how to pin one.
Speed. The speed setting (0.7 to 1.2 on ElevenLabs voices) is an API field; the composer has no slider yet, so a take that runs long is rewritten or trimmed.
Exact timing. Music comes back longer than the clip, ours two minutes for five seconds of picture, and the beat lands where the model puts it. You mark in and out to the clip, not the other way round.
One effect per run. A prompt with three sounds in it returns one blend.

What it costs

Speech bills per character with a minimum for a short line; music and effects are one charge per request. The clip is priced by its length. Live rates for the models used in this guide:

ElevenLabs v3Every plan
Per 1,000 characters
Standard 6 credits
Lyria 3.5Every plan
Per generation
Standard 5 credits
ElevenLabs SFXEvery plan
Per generation
Standard 4 credits
Seedance 1.5 ProPro and up
Per 5s clip (audio on by default)
480p 7 credits · 720p 15 credits · 1080p 33 credits

Read live from the catalog when this page loads. Every price is shown before you generate. See every rate.

The Studio export is not billed. New accounts start with 350 free credits.

Try it yourself

The source clip and the three prompts

Fragrance clip, silent, 5 s, 1280x720MP4 video · 1280x720 · 1.2 MBDownload

Upload the clip to your Library, run the three prompts above on the recommended models, then add all four at the playhead in Studio.

Script into Voiceover on ElevenLabs v3, any named voice; ours was Roger.
Music prompt into Music on Lyria 3.5. Expect a track longer than the clip.
Effect prompt into Sound effect on ElevenLabs Sound Effects v2.
Timelines, New timeline, then in Studio switch the left panel to Library: clip first with its +, then the music through Mark In and Mark Out in the Source preview, then the narration and the chime at the playhead. Export video.

Questions and answers

Can I make the voice, the music and the effect in one place?: Yes. The audio composer has a Voiceover, a Music and a Sound effect task, each with its own models, and every result is an audio asset in your Library.
How is audio priced?: Speech is billed by the characters in your script, with a small minimum; the button shows the price as you type. Music and sound effects are one flat charge per request, shown on the button before you generate.
Does the narration line up with the picture automatically?: No. You place it. In Studio each audio asset is its own item on an audio lane, so you slide the take to the frame you want and trim the music to the clip.
Should I just let the video model make the sound?: Some models can, and it is quick. You lose the ability to change one sound without rerolling the picture. Render silent when you want to control the mix.
Does exporting from Studio cost credits?: No. The MP4 render is not billed; only the generations that made the clip and the sounds were.
Can one prompt make two sound effects?: It returns one blend. Make each cue its own generation, then place them separately.

Give your clip a soundtrack

Open the audio composer

Audio

The same voice in every video

Save a catalog voice on a character, read every line with Speak as, and lay a music bed and a sound cue under it in Studio. One narrator, three clips, made for real.

September 24, 2026 · 10 min read

Channels

A faceless narrated channel: the tools, the workflow, the next episode

Script, one saved narrator voice, b-roll in a locked look from the video composer, the cut in Studio, and the NOLGIA Agent for the next episode. Two short episodes made for real.

September 24, 2026 · 11 min read

Credits

How credits work, and what happens when a render fails

The price is on the Generate button, the job holds it while it works, and the hold becomes a charge or a refund. Here is when credits come back after a failure, a refusal or a cancel.

September 24, 2026 · 6 min read