Cast an AI voice the way a director casts an actor: for who the character is, meaning their age, home, texture, pace and how they sound when things go wrong, then keep that voice for the whole series. In Agent Social every character gets one, and the lead can have yours.
What is an AI voice actor?
An AI voice actor is a generated voice cast for one character and kept for every line they ever say, so after a scene or two the voice simply is that person. Treat it as casting, the most underrated job in a film. Cast badly and every scene the character is in feels slightly off, however good the picture; cast well and the voice does half the acting, because viewers decide whether to trust someone within a line or two. You can cast a voice described in words, or clone a real one from a recording you are allowed to use, your own included. Once cast, the voice speaks every line in every clip, with lips moving in sync and the sound of the scene underneath. The three leads of OVERCLOCK, my kaiju series, were cast for the first episode, and none of them has been recast since.
One voice reads everything
- A single narrator reads every part
- Characters are described, never heard
- Every scene carries the same tone
- Eyes closed, you can’t tell who is who
A cast with a voice per character
- Each character speaks their own lines
- Lips move with the words, in sync
- Each voice kept for the whole series
- A narrator only where the story needs one
How do you choose the right AI voice for a character?
Choose the voice from five facts about the character and let it follow from them: age, home, texture, pace, and how they sound under pressure. Age and home set the base: a retired fisherman from the Konkan coast does not sound like a young trader in Singapore. Texture is the grain, gravelly, breathy, bright or husky, and pace is how fast they think out loud. Pressure is the one people forget, and it matters most, because drama is people under pressure: does she get louder, or quieter and clipped? Then try casting against the obvious. The biggest man in the room with the softest voice, or a villain who never raises his, stays with viewers longer than the growl they expected. Put the five facts in one sentence and use it as the casting brief when you make the character. The casting sheet below has six ready to copy.
The gravelly old fisherman
Slow, rough and kind: the voice that makes a warning sound like wisdom.
“Give Anand a voice: a man in his seventies from the Konkan coast, gravelly and slow, who never raises his voice, even in a storm.”
The calm ops-room officer
She gets quieter as things get worse, which is scarier than shouting.
“Give Major Rao a voice: a woman in her forties, low and even, clipped and precise, who gets quieter when things go wrong.”
The fast-talking best friend
Speed is the joke: she thinks out loud and starts the next sentence before finishing the last.
“Give Zoya a voice: late twenties, bright and quick, a Bandra accent, laughing in the middle of her own sentences.”
The villain who never shouts
A soft, polite voice makes every threat land harder than a roar.
“Give Mr Kale a voice: a man in his fifties, soft, smooth and very polite, never hurried, with a faint smile in every line.”
The grandmother who remembers
Unhurried, with a smile in it: the story sounds like something she lived, not something she read.
“Give the narrator a voice: a woman in her seventies from Pune, unhurried and fond, telling the story as something she remembers.”
The chai-stall owner from Lucknow
Everyday Hindi mixed with English, an unhurried manner and a tease for every customer.
“Make this series speak Hindi, and give Pappu a voice: a man in his fifties from Lucknow, warm and unhurried, teasing every customer.”
When should you cast your own voice, and when someone else’s?
Cast your own voice when the role is close to you, and someone else’s when it is not. Two kinds of role suit you best. The first is a lead who is a version of you: your age, your city, your temper, so every line sounds like you because it is. The second is any role on a channel whose viewers already know how you sound, where your voice becomes the show’s signature. Cast someone else for anyone decades older or younger than you, from a town you have never lived in, or with the opposite of your nature: you will hear yourself acting, and so will everyone else. Villains are the classic trap, because a nice person playing menace sounds like a nice person trying. And avoid voicing two main characters yourself, or viewers will hear one person twice. Once your voice is cast, keeping a cloned voice sounding like you covers the rest.
How do you cast a narrator’s voice?
Cast the narrator for the listener, not for a scene: the narrator talks to the viewer, so the voice should sound like someone they would happily sit with for the whole story. A bedtime fable wants warm, close and unhurried. A mystery wants dry and even, a little amused. A recap wants quick and clear. Your own voice is the strongest choice when the story really is yours. Keep the narrator unlike every character, so nobody wonders who is talking, and give them fewer lines than you think: anything the characters can play belongs in their own scenes, which an AI video with dialogue shows how to write. If you plan to earn on YouTube, its policy, as of November 2026, says videos that only read out material you did not create cannot earn, so let the narrator tell your own story. Making AI story videos covers when a story needs one at all.
How do you make each AI character sound different from the others?
Cast for contrast: put the main voices side by side and make sure no two share the same age, pace and texture. A scene works when the audience knows who is speaking before they look, and two voices of the same age at the same speed blur into one. Change at least two things between any pair: an old, slow voice against a quick, young one, a gravelly voice against a bright one. Contrast lives in the words too. Give one person long, careful sentences and another single words, and they stay apart even when both whisper. For a Hindi cast, contrast matters even more, because everyone shares one language: give the auntie, the cousin and the shopkeeper different ages, speeds and home towns. The Hindi web series guide shows how a Hindi series is set up, and Hindi AI movies shows what to ask for.
A face tells you who someone is. A voice tells you whether to believe them.
How do you direct an AI voice actor’s performance?
Direct the performance through the scene, not the voice: say what the character is doing, who they are talking to and what they want, and the delivery follows. A line said while running, while eating or into a radio over gunfire sounds different without the voice changing at all. Take three words from OVERCLOCK Episode 5: “Holiday’s cancelled, sir,” said by a soldier as he hands Kevin a radio. Shouted across a street, they would be an alarm; said as he passes the radio over, they land as dry understatement, in the same voice. The other half of a performance is the writing. Frightened people speak in fragments, proud people in full sentences, and someone hiding something answers a question nobody asked, so fix the line before you blame the voice. In Agent Social you read every scene before it is filmed, and you can rewrite how anyone talks in plain words.
How do you hold a table read for an AI cast?
Hold a table read by hearing the whole cast together, not one voice at a time: play every main character’s audition back to back, in the order they first speak, with their pictures in front of you. A voice that sounds fine alone can fail next to the others, so listen to the ensemble. Play the two most alike one after the other; if you hesitate over who is who, recast one of them. Check the pecking order: the mentor should sound older than the student, the boss surer of herself than the clerk. Then imagine each one on the hardest line they will ever say, the shout, the confession, the bad news, and ask whether you believe it. Recast anyone who fails now, before a single scene is filmed with them. A table read catches what one audition hides, like two friends who sound alike or a villain too gentle for his worst scene. After that, keeping each voice the same is its own job.
Write who they are
Age, home town, job, and what they want in this story. The voice grows out of the person.
Cast the voice
Texture, pace and pressure in one sentence, or a clone of a real voice you are allowed to use, yours included.
Hold a table read
Hear every audition back to back beside each face. Recast any voice that blurs into another.
Write lines that fit
Rewrite any line the character would never say, then carry the same cast into every episode of the series.
How much do AI voice actors cost?
There is no per-actor fee: every line is spoken inside the clip. Nobody is booked for a session, paid by the hour or synced afterwards, because each clip comes back with the lines, the lip movement and the sound of the scene already in it. So a scene with five people talking is priced by the finished minute, exactly like a scene with one, and a big cast adds casting work, not a higher price per minute. What adds to the total sits around the edges. Designing each voice uses some credits, as writing and pictures do, and a narrator’s voiceover is made on top of the video, so a narrated film costs a little more than one carried by dialogue. It is pay as you go, with no subscription, and credits never expire. The pricing page has the rates, what an AI episode costs prices a whole episode, and how music, voices and subtitles fit together covers the finish.