How to keep an AI character’s voice the same in every clip and every episode

To keep a character’s voice the same in Agent Social, set their voice once in their Voice panel and audition it before you render anything. From then on, every line that character says, in every clip and every later episode of a series, is spoken in that voice. You never pick a voice per clip.

In OVERCLOCK, my AI kaiju series, Deep, Nora and Otto sound the same in Episode 4 as they did in Episode 1. Here is how to set that up, and what to do when a voice comes out wrong.

Why do AI voices drift between clips?

When a voice is left for each clip to invent, every clip invents a slightly different person: a little older here, a different accent there. A face can look right while the voice gives the drift away. The fix is the same as for faces: decide the voice once, and use that decision everywhere the character speaks.

Step 1: give the character a voice

Open the character and go to their Voice panel. There are three ways to give them a voice:

• Describe it in words: age, tone, accent and pace, for example “a woman in her thirties, warm, a little dry, a Mumbai accent”.
• Record your own voice and set it, which is how you can star in your own film with your own voice.
• Upload a recording you have the right to use.

Whichever way you choose, play the audition before the voice goes anywhere near a clip. This is the moment to hear whether the voice fits the face. Changing it now costs nothing; changing it after twenty clips means re-rendering every clip you want in the new voice.

Step 2: write down how they speak

Give every character a short written voice line next to their look: age, tone and accent. Otto’s is gravelly, gentle and unhurried. That one line keeps him sounding like Otto in every scene, because every line he says is written to match it: short sentences, nothing rushed, a dry joke now and then.

How a line is said can still change from scene to scene: quietly, half laughing, shouting over a helicopter. That is acting, not a new voice. The person stays the same; only the mood moves.

Step 3: render one clip and listen

Render the first clip with the character in it and listen with your eyes closed. Does the voice match the person you see? Is it the same person who speaks in the next clip? Clips render in story order, so the first speaking clip is your test before the rest of the film commits to the voice.

What if a voice sounds unreal in a clip?

Sometimes a set voice makes a character sound a little unreal in one scene. Each character has a switch that lets the video model voice them itself instead. Their lines stay the same; only clips rendered after you flip it change. Use it when the set voice fights the scene, and flip it back if the character starts to drift.

Does the voice carry into the next episode?

Yes. A series keeps its cast, so Episode 2 uses the same pictures and voices as Episode 1. When you start the next episode, the characters who carry over are reused, not made again, so nobody comes back sounding like a stranger.

What about people who only speak once?

Someone who appears once, like the fishing skipper whose mayday crackles over a helicopter radio in Episode 1, does not need a Voice panel. Describe them in the scene instead: “a weathered fisherman, shouting over the wind”. Give a set voice to anyone who speaks in two or more clips, so they sound like the same person every time.

Starring with your own voice

If you play a character yourself, record your voice once and set it on that character, the same way you would use your own photos for their face. Then ask for lines the way you actually talk. A recorded voice reading words you would never say still sounds like someone else.

Accents, and characters who speak Hindi

An accent is part of the voice, so set it in the Voice panel, not clip by clip. In a Hindi series, each character’s voice is designed with an Indian regional accent of their own, such as a Mumbai local or someone from Delhi, and their English words carry the same accent. Exactly how closely a given accent matches its region has not been tested yet, so listen to the first clips and adjust the voice if it drifts.

Can I change a voice later?

Yes. Change it in the Voice panel. Clips already rendered keep the old voice, and clips rendered after the change use the new one. For a voice change in the middle of a film, re-render the clips the character speaks in so the whole film sounds like one person.

A checklist for consistent voices

• Every speaking character has a voice set in their Voice panel.
• You have played every audition, and each voice fits its face.
• Each character has a written voice line: age, tone, accent.
• You listened to the first speaking clip before rendering the rest.
• Any voice change is followed by re-rendering that character’s clips.