Skip to content
All tutorials

Film 7 of 9 · Cast · 4:30

A voice that is a character

A voice designed from a description, never a real person’s, cast on its card, and a line directed take by take.

Transcript

A character's voice is part of who they are. Design it from words, never from a real person, and cast it once, so every line sounds like the same character. In this video: a voice designed from words, cast on its card, and a line directed take by take.

The lighthouse in Every Nine Seconds: we hear him long before we ever see him. We designed his voice at ElevenLabs, from a description: an old keeper, low, dry, unhurried, reading his logbook, never theatrical. No real person's voice: designed, never cloned. It made three previews.

We kept one, and saved it to the account. In Kiju, a voice is cast on the character's card, like a look: the lighthouse's card names his. When you make his lines, you choose it from the Cast.

In the Lab, each of his lines is its own take. Search the prompts for his words, and show it: short flat sentences, written for him. Its details show the voice, the model, and the cost: under a cent a line.

Now, yours. Snail Express's bible says the narrator talks, and Shelby never does. So the narrator needs a voice. At ElevenLabs, open Voice Design and describe him: a man in his fifties, warm and gentle, like a favourite uncle telling a bedtime story; cheerful, never theatrical. Give him a few lines of your show to read. Listen to the previews, and save the one you choose to your account.

Back in Kiju's Cast, choose New character: the Narrator, who tells the story. Not a real person. Add him to the cast, then choose his voice: Kiju lists the voices of the ElevenLabs account you connected, in Connections.

Then shot 2.1, in the Lab: a Voice take. Open the voice menu, and choose The Narrator under Cast. Write his line, and direct it with tags in brackets: warm, a pause. The price is on the button, in cents: this line costs about a cent and a half. Generate.

One line per take: to try another reading, change the direction, not the words. Softly, a whisper for the ticks, warm again at the end, and one word in capitals for stress. Generate again.

The first is warm and steady: a storybook. The second whispers the ticks, then jumps back to full voice: too secretive for a cheerful show. Star the first, and discard the second: it's set aside, with its cost. Don't Keep a voice take for 2.1: Keep picks the shot's one take, its picture. A starred line waits for your edit.

Pause here, and make yours: a voice from a description, cast on the card, one line per take, directed with tags. Next time: notes on the time, in the cutting room.

The film is in English, with English captions. Filmed on the Kiju desk, version 0.1.0. The images, the video and the voices in the films are AI-generated.