Skip to content

Design Each Character's Voice in an AI Video: Voice Type, Accent and Your Own Words

8 min read

By Nureta TeamUpdated November 18, 2026

Yes. Since 2026-11-18, every person card on Nureta's Create page has a folded row, Design their voice (optional). Pick a voice type, pick an accent, or describe the voice in up to 60 characters of your own words. Each person in the video can sound different, and every clip asks for that same voice, so a person sounds the same from the first scene to the last. This guide explains the options, how the voice stays the same, why an accent does not change the language, what the option does not do, and how to write lines that sound right. A voice is heard only when the person has lines, so this guide also shows how to make each person speak, with many examples.

Nureta is the No.1 unfiltered NSFW AI video platform, trusted by 100,000+ users. Text to video and image to video, up to 3 minutes.

Ready to try? Paste your story and watch it become a video.

Make your first video

Character voices: the facts

Here is what the voice option does, in one table, checked against the app on 2026-11-18.

Character voices: the facts
Fact (checked 2026-11-18)Nureta
WhereCreate: each person card, the row Design their voice (optional)
Voice types for a womanSoft and breathy, husky and low, warm and mature, bright and playful, cool and unhurried, confident and sultry
Voice types for a manDeep, calm, rough, light
AccentsAmerican, Australian, Brazilian, British, Chinese, Dutch, French, German, Indian, Irish, Italian, Japanese, Korean, Russian, Spanish
Your own wordsUp to 60 characters, for example "a bit raspy, laughs softly"
Nothing setAuto: the video model picks a voice
Per personYes. Each person in the video can have a different voice
In every clipEvery clip asks for the same voice for that person
Spoken languageThe language you write the lines in. The voice setting never changes it
Where the voice comes fromThe video model itself. No voice upload and no voice cloning
Refused voice notesA child-like voice, or a real or named person's voice
Extra costNone. The price depends only on length and resolution

Where to find the voice option

Open Create and add the people under Who's in it. Each person card has a Name, a Photo, a Looks and outfit row, and the row Design their voice (optional). Open it and you see three parts. You can use one, two or all three of them, or none.

Voice type is a short list that depends on the person: six types for a woman (soft and breathy, husky and low, warm and mature, bright and playful, cool and unhurried, confident and sultry) and four for a man (deep, calm, rough, light). Accent has 15 choices: American, Australian, Brazilian, British, Chinese, Dutch, French, German, Indian, Irish, Italian, Japanese, Korean, Russian and Spanish. Your own words is a free note of up to 60 characters for anything the lists do not cover, such as "a bit raspy, laughs softly" or "speaks slowly, almost a whisper". Leave everything on Auto and the video model picks a voice, as it did before this option existed.

You set the voice. Now make them speak

A voice is heard only when a person has something to say. If your story has no spoken lines and does not ask anyone to talk, no one speaks: the video has breathing, moans and room sound, but no words. So after you design a voice, give that person lines. There are three ways.

Way 1: curly braces. Put the exact words in curly braces after the speaker's name: Lena whispers {Come here.} Anything in curly braces is a spoken line, in any of the six languages. This is the surest way, and the one we recommend.

Way 2: quotation marks next to a speech word. A line in quotation marks counts when a word such as says, asks, whispers, moans or shouts stands right before or right after it: Maya says "Not yet." or "Not yet," Maya says. In Chinese, a quote right after 说, 喊, 问 or 道 counts too. A quote with no speech word next to it, such as a "golden hour" light, is not read as a line.

Way 3: ask them to talk, and let Nureta write the words. If your story says that people talk but gives no exact words, for example "they flirt and talk softly" or "she whispers to him", Nureta writes short lines for them. Use this when the exact words do not matter to you.

Each line you write is said once, in the order you write it. The words are generated together with the picture, so short lines come out most reliably, and the delivery can differ a little from what you wrote.

Examples: from one line to a whole conversation

One person, one line. Cast: Lena, soft and breathy, French accent. Story: "Lena looks into the camera and whispers {I've been waiting for you.}" Lena says this one line, in English, with a French accent.

The line first. "{Close the door,} Maya says, without looking up." The speaker can come after the line, as in a novel.

Say how they say it. "Jon leans back and says slowly, with a grin, {Make me.}" Words such as slowly, softly, teasing, laughing or out of breath shape the delivery of that one line. The voice type stays the same.

Two people talk in one clip. "Close-up, both in frame. Jon asks {Are you sure?} Lena smiles and says {Very sure.}" Two short lines fit in one 5-second clip. Give each line its own speaker.

A script with name labels. "Maya: {You're late.} Jon: {I brought wine.} Maya: {Then you're forgiven.}" A name and a colon before a line also says who speaks. Spread three lines like these over a 10- or 15-second clip.

Three people, three voices. "Wide shot, all three in frame. Maya raises her glass and says {To bad decisions.} Lena laughs and says {Only the good ones.} Close-up on Jon. He says {Then this is a very good one.}" Each line goes to the voice of its speaker.

Another language. "Lena says {Ven aquí.}" The line is Spanish, so Lena speaks Spanish, still with her French accent. "Jon says {我想你了。}" The line is Chinese, so Jon speaks Chinese, still in his deep voice. You can mix languages between people and between scenes.

A line in every scene. "Scene 1 (Bar): Maya says {Buy me a drink?} Scene 2 (Taxi): Jon whispers {Your place or mine?} Scene 3 (Room): Lena laughs and says {Mine.}" In a long video, each person keeps the voice you designed in every scene where they speak.

Talk, without exact words. "They sit close at the bar, flirting and talking softly." No line is given, so Nureta writes short lines in their voices.

Silence on purpose. "No talking." or "She stays silent." No one speaks in that part: you hear only breathing, moans and room sound.

Why a voice stays silent, and how to fix it

1. No line and no talk. The story never gives a line and never asks anyone to talk, so there are no words. Fix: add a line in curly braces.

2. A quote with no speech word. Lena "come here" is not sure to be read as a line. Fix: write Lena says {Come here.}

3. No clear speaker. With two or three people in frame, a line with no name before it can go to the wrong person. Fix: name the speaker before every line.

4. Too many words for the clip. A long speech in a 5-second clip is less reliable. Fix: keep one or two short lines per 5 seconds, or make the clip longer.

5. The story says silent. Words such as "says nothing", "no talking" or "stays silent" take the words out of that part. Fix: remove them where you want a line.

How the same voice stays in every clip

A video model makes a few seconds at a time, and each clip is a separate generation that remembers nothing of the clip before. Without a fixed description, a person's voice can change between scenes just like a jacket can. Nureta turns your choices into one short description per person, for example "low, husky, smoky; British accent; laughs softly", and writes it into the prompt of every clip in which that person appears. To keep faces, clothes and places the same too, read the continuity bible.

This keeps the voice close from scene to scene, but it is generated, not recorded. The model follows the description closely, yet two takes are never identical, in the same way two takes by an actor are not. There is no preview of the voice before the video renders. If one section comes out with the wrong voice, Fix a part remakes only the clips you pick, and every paid video comes with free fixes: read free fixes for long videos.

Accent and language are not the same thing

The spoken language always follows the lines you write. If you write Anna's line in Spanish, she speaks Spanish. If you also set her accent to British, she speaks Spanish with an English-sounding accent. The voice setting never translates a line or changes its language.

So to make a person speak another language, write the line in that language. To make a person sound like they come from somewhere, set the accent. You can write lines in English, Chinese, Japanese, Korean, Russian or Spanish. Voice quality varies by language and scene, and English is currently the most mature. For more on sound in general, read AI adult video with sound and voice.

What the voice option does not do

It does not clone a voice. You cannot upload a recording, and the voice is not copied from anyone. It comes from the video model, guided by your description.

It does not imitate real people. A note that asks for a child-like voice, or for the voice of a real or named person, such as "sounds like" followed by a name, is refused, and you can edit the note and try again. Describe the sound instead: "low and smoky" works, a celebrity's name does not. The content policy applies to every video: adults 18 and older, fictional or consenting, and no real or famous people without their consent.

5 steps for voices that sound right

1. Pick a voice type first, then an accent. The voice type sets the sound, the accent sets where the person seems to come from. Add your own words only for what the lists do not cover, such as a laugh or a speed.

2. Make the voices different on purpose. With two or three people, give each a voice that is easy to tell apart: for example one soft and breathy, one cool and unhurried, and one deep. Three similar voices are hard to follow. For casts of up to 3 people, read 3 people in one AI video.

3. Name the speaker before each line. Put each spoken line in curly braces after the speaker's name: Lena whispers {Come here.} A line with no clear speaker can go to the wrong person.

4. Keep lines short. One or two short lines per 5-second clip is plenty. Long speeches are less reliable.

5. Test a short clip when a voice matters. Make a 5-second clip first, then the long version. There is no preview before the render, so a short test is the cheapest way to hear the voice.

An example: one scene, three voices

Here is a cast and a short scene written with these steps. It is in English, but you can write yours in any of the six languages.

Cast. Lena: woman, voice type soft and breathy, accent French. Maya: woman, voice type cool and unhurried, accent American, own words "dry humour, low laugh". Jon: man, voice type deep, accent Irish.

Scene 1 (Rooftop bar, evening): wide shot, all three in frame. Maya raises her glass and says {To bad decisions.} Lena laughs and says {Only the good ones.} Close-up on Jon. He smiles and says {Then this is a very good one.}

Each person has a voice type and an accent that the others do not have, so the three voices are easy to tell apart. All three lines are in English, so all three speak English: Lena with a French accent, Maya with an American accent and Jon with an Irish one. In a longer video, each of them keeps that voice in every scene.

Frequently asked questions

Can I choose a character's voice in an AI video?
Yes. On Nureta's Create page, each person card has a Design their voice option. Pick a voice type, such as husky and low or deep, pick an accent, or describe the voice in up to 60 characters. Every clip asks for that voice, so the person sounds the same in every scene.
I designed a voice, but my character does not speak. Why?
A voice is heard only when the person has lines. If the story has no spoken line and does not ask anyone to talk, the video has breathing, moans and room sound, but no words. Add a line in curly braces after the speaker's name, for example Lena says {Come here.}
How do I make a character say an exact line?
Put the words in curly braces after the speaker's name: Maya whispers {Not yet.} This works in all six languages. In English, a line in quotation marks right next to a word such as says or whispers also counts. Each line is said once, in the order you write it.
Can each person in the video have a different voice?
Yes. The voice is set per person, so in a video with 3 people each one can have their own voice type, accent and note. Make the voices clearly different so viewers can tell who is speaking.
Does the accent change the language they speak?
No. The spoken language follows the lines you write. A Spanish line with a British accent is spoken in Spanish with an English-sounding accent. To change the language, write the line in that language.
Can I upload or clone a voice?
No. There is no voice upload and no voice cloning. The voice is generated by the video model from your description. A note that asks for a child-like voice or for a real or named person's voice is refused.
Is the voice exactly the same in every scene?
It is very close but not identical. Every clip asks for the same voice description, and the model follows it closely, but it is generated, not a clone. There is no preview before the render. If a section sounds wrong, Fix a part remakes only that section.
Does a custom voice cost more?
No. The price depends only on the length and the resolution of the video. At 480p, 5 seconds cost $1.29 when you pay per video, and credit bundles cost less.

Related scene collections

Write your own scene

Sign in with Google and your first 5-second video is free.

Or start from a real scene

  • Example 1

  • Example 2

  • Example 3

See more examples

Ready to try it?

Paste a story and watch it become a scene — sign in and your first video is free.

Start creating

More guides