# Making audio

> Making voice-overs, music and sound effects from the Composer or by asking the Assistant, what each costs, and playing, downloading and finding them on the Canvas and in your library.

Choose what to make from the mode button at the start of the prompt (see [Talking with the Assistant](/create/assistant/)). **Audio**, under Image and Video, opens its three jobs, and you always choose which one: the words never decide it. Once one is chosen, the mode button shows its name.

- **Voice-over**: write the words to speak, and it is read aloud in the voice you choose.
- **Music**: describe the music, such as "a slow, warm piano theme with strings".
- **Sound effect**: describe the sound, such as "rain on a tin roof".

There are three ways to start:

- **The Composer's Audio**: open the mode button, choose **Audio**, then **Voice-over**, **Music** or **Sound effect**.
- **Create audio**, on an empty Canvas, asks the same, and the Composer is in that mode, ready for you to type, with a blank card waiting where the sound will land (see [An empty Canvas](/work/canvases/#an-empty-canvas)).
- **Right-click** empty space on a Canvas and choose **Add**, **Audio**, then the job: the Composer is in that mode, ready for you to type, with a blank card where you right-clicked, and the sound lands there. Pressing the Canvas takes the blank card away; the sound still lands where you asked.

Press **Make voice-over**, **Make music** or **Make sound effect** (or Enter) to send it. On a Canvas, its place appears at once and glows while it is made, like an image's, then becomes its card. In a Space's library, it is made into the library. Until it is ready, **Stop** in the Composer cancels it at no cost (see [Stop](/create/assistant/#stop)). Audio is made from your words alone: images you've attached stay attached for your next image or video, and aren't sent. Audio isn't made from an image's editor.

## Asking the Assistant

In **Ask**, you can ask the Assistant for audio as you ask it for an image: "Make a voice-over that says: Welcome to the gallery", "I need the sound of a door creaking, about three seconds", "Write some calm piano music for the intro, no vocals". It works out which job you mean from what you say, and what you said about it: a voice by name, a length, **Loop** or **Instrumental**. Nothing in Otherspace guesses the job from your words: the Assistant chooses it, and its card says which it chose.

- A suggestion comes as a prompt card that names the job and what it will be made with ("Voice-over · Aria", "Sound effect · 3s · Loop", "Music · 1:00 · Instrumental"). **Make it** makes it as it is. **Use** puts it in the Composer in that job, with those settings, to change before you send.
- A clear request ("Make a voice-over that says…") is made straight away, like an image, with no "are you sure?". The exact price is worked out before it is made, from its words or its length, and it is charged like anything you make from the Composer. What you haven't said is the job's setting in the Composer, or the model's default.
- For a voice-over, the Assistant's prompt is exactly the words that will be spoken.
- What it made shows in the conversation with its job and length: press play there to listen (one sound plays at a time, anywhere on the page), or **Download** to save it. On a Canvas it is placed as its card; in a Space, it is made into the library.

The Assistant makes audio from a Canvas, a Space or Home (where you choose the Space to make it in), on a phone too, but not in an image's or a document's editor. It never sends the images you've attached with audio.

## Settings

Each job has its own settings, in **Settings**, the sliders button beside the [cost meter](/account/cost-meter/) (see [Settings](/create/images/#settings)). Each job remembers its own.

- **Voice-over**: its **Voice**, one of ElevenLabs' 21 preset voices (Rachel unless you choose another). Each voice in the list says how it sounds and what it suits, such as "Warm, mature man, British · storytelling", and has a play button to hear it introduce itself before you choose. Previews are free, one plays at a time, and closing the list stops it. Then choose how it is read: **Stability** (higher is steadier, lower more expressive) and **Similarity** (how closely it keeps to the voice), each from 0% to 100%. A voice-over can be up to **5,000 characters**; near the limit, the count shows beside the send button, and past it, the send button waits until it is shorter.
- **Music**: its **Length**, from 3 seconds to 10 minutes, and **Instrumental**, for music without singing.
- **Sound effect**: its **Length**, from half a second to 30 seconds in tenths of a second, and **Loop**, for a sound that repeats seamlessly.

Unless you choose a model in Settings, a voice-over uses **ElevenLabs v4**; **ElevenLabs v4 Turbo** is the fast option, at half the price. Music uses **ElevenLabs Music v2.5**, and sound effects **ElevenLabs Sound Effects v3** (see [Models](/models/)). You can also end a prompt with `--model eleven-sfx`, say, to use that model for it.

## What it costs

The exact price shows on the [cost meter](/account/cost-meter/) before you send, and follows your words and settings:

| Job | Priced by | Price |
| --- | --- | --- |
| Voice-over, ElevenLabs v4 | The characters it speaks | $0.11 per 1,000 characters |
| Voice-over, ElevenLabs v4 Turbo | The characters it speaks | $0.055 per 1,000 characters |
| Sound effect | Its length, each started second whole | $0.003 a second |
| Music | Its length, each started minute whole | $1.10 a minute |

Music is billed per started minute: 30 seconds is billed as one minute, and 1 minute 30 seconds as two. A sound effect is billed per started second: 2.5 seconds is billed as 3. Settings say so under the length. Something that can't be made costs nothing (see [Usage and prices](/account/usage/)).

## Playing it

On the Canvas, a sound is a card: its name and its length at the top, and its waveform below. Hover it, or select it (tap it on a phone or tablet), and a play button shows at the start of its waveform: press it to play, and again to pause. Press or drag along the waveform to play from that point; with the waveform focused, the arrow keys move five seconds, and Space plays or pauses. What has played is solid, what is still to play is dim, and a thin red line shows where it is. Until its waveform has been worked out, a dotted line holds its place; it still plays.

One sound plays at a time: starting another, anywhere on the page, pauses the one playing. Playing is yours alone: pressing play never plays anything for anyone else. When it leaves your view, or you switch to another tab, it pauses where it is, and waits for you to play it again.

It is an ordinary node: move it, resize it (its waveform fills whatever size it is), copy or delete it. If it can't be made, it turns red like an image does, with **Retry**. Select it and its properties' **Details** show its type, length, file size and the model that made it; its **Metadata** shows the generation that made it, and **Reuse prompt** puts its words back into the Composer in the same job.

## Downloading it

Right-click a sound and choose **Download original** to save its file exactly as it was made.

## In your library

In a Space's library, a sound's card shows its waveform and its length; hover it to play it (on a phone, its play button is always there). Its details play it too. Open it, or press Space, for a closer look. Choose **Audio** under **Kind** in the filter to see only sounds. **Add to Canvas** places it on the Canvas as its card.

A step of a [prompt template](/create/space-knowledge/) can make a voice-over, music or a sound effect too.

Not available yet: uploading or recording audio, a sound as a reference, and sound for videos.
