Give Every AI Character Its Own Voice
Why AI characters change voice between clips, and how to keep one consistent voice per character: design it from a description or upload a short recording.
To give an AI character a consistent voice, set the voice once and reuse it in every clip. In the AI Drama Series Generator you can design a voice for each character from a short description (free, 3 tries) or upload a 2 to 15 second recording you have consent to use. Every episode then uses that voice. This works on the Seedance models, for up to 3 speaking characters per episode.
Disclosure: VdoBloom is our product.
A drama series asks viewers to follow the same people across many episodes. They notice when a face changes. They also notice when a voice changes. If the hero sounds deep and calm in episode 1 and thin and fast in episode 3, the spell breaks. This tutorial explains why that happens and how to stop it.
Why voices change between AI clips
Video models that speak make up a voice for each clip. They look at the scene, the character and the line, and invent a voice that fits. Nothing tells the model what this person sounded like last time. So each new clip can bring a slightly different voice, or a very different one.
Faces have the same problem, and the drama tool fixes it with reference images: a character sheet for each person and an image for each location. Voices need the same kind of reference. That is what voice design gives you: one voice per character, stored once and sent with every episode.
Step 1: Add your cast
Each series can have up to 4 characters. For each one, add a photo or describe the face and hair plus the outfit. Each character then gets a reference: a generated character sheet for 5 credits, or your own photo for free. The voice step comes next.
Step 2: Choose how to make each voice
For every character, you have two options.
Option A: Design a voice from a description
Type a short description of how the character should sound. The tool makes a voice from it. This is free, and you get 3 tries per character, so you can adjust the description if the first result is not right.
If you leave the description blank, the tool still makes a voice. It uses the character's face text. If there is no face text, it uses the character's photo. This is a fast way to get a voice that fits the look, but a written description gives you more control.
Option B: Upload a recording
Upload a clip of a real voice, 2 to 15 seconds long. The character will then speak in a voice based on it. You must have consent from the person in the recording. Use your own voice, or the voice of someone who has agreed to it. Do not upload a recording of someone who has not said yes.
For best results, use a clean clip: one speaker, little background noise, normal speaking pace.
Step 3: Every episode uses that voice
Once a character has a voice, you do not need to do anything else. The AI writes the season plan, including dialogue, and every episode you make uses each character's voice for their lines. The characters speak with native lip-sync in your chosen language.
This is the main gain. You set the voice once at the start, and it stays the same from the first episode to the last.
Limits to know
- Seedance models only. Designed and uploaded voices work with Seedance 2, Seedance 2 Fast, Seedance 2 Mini and Seedance 2.5. Wan 2.7 makes its own voices, so it does not use the ones you set.
- Up to 3 speaking characters per episode. You can have 4 characters in the cast, but in any one episode, up to 3 speaking characters get their voice. Plan scenes so that no more than 3 people talk in the same episode.
- Finished episodes keep their old voice. If you change a character's voice later, episodes you already made do not change. Only new episodes use the new voice. Decide on voices before you make episode 1, or be ready to remake earlier episodes.
Tips for good voice descriptions
A good description is short and concrete. Think about what you would tell a casting director. Cover these points:
- Age and gender, for example "woman in her 30s".
- Pitch: low, medium or high.
- Pace: slow and careful, or quick and nervous.
- Texture: warm, husky, crisp, soft, rough.
- Attitude: confident, shy, cold, playful.
Avoid naming real people or famous actors. Describe the sound instead.
Example descriptions
| Character | Voice description |
|---|---|
| Cold CEO (CEO drama) | Man in his late 30s, low and calm voice, slow pace, few words, quiet authority, never shouts. |
| Bright heroine (Romance) | Woman in her mid 20s, medium-high pitch, warm and quick, a little breathless when nervous. |
| Scheming stepsister (Revenge) | Woman around 30, smooth medium voice, sweet on the surface with a sharp edge, slow when threatening. |
| Old mentor (Fantasy) | Man in his 70s, deep and slightly rough, slow and patient, soft but firm. |
| Kid sidekick (Kids episode) | Child around 9, high and bright, fast and excited, laughs easily. |
Make voices easy to tell apart
If two characters often talk to each other, give them clearly different voices. Pair a low, slow voice with a higher, quicker one. Viewers should know who is talking even if they look away.
Use your 3 tries well
If the first try is close, change one thing at a time. "Lower pitch" or "slower pace" is better than rewriting the whole description. That way you can tell which change helped.
Write for the scene, not just the person
Think about where the character speaks most. A voice that shouts across a busy street needs a different description from one that whispers in an office at night. If most of their lines are quiet threats or soft confessions, say so: "speaks softly, rarely raises her voice." The description should fit the scenes in your season plan.
Match voice to story language
Your series has one spoken language, chosen at setup. Think about how the character should sound in that language. See our guide to AI short drama lip-sync in 8 languages for the supported list.
Other places to use voice design
If you want a voice outside a drama series, for a voiceover or a single clip, try AI Voice Design on its own. For longer projects that mix several tools, AI Studio can help plan them.
What a season costs
Voice design from a description is free. The costs in a series are the references and the episodes. Each episode has a 5-credit storyboard, then the video. For example, 5 episodes on Seedance 2 Mini at 720p, 15 seconds each, is 5 x (5 + 50) = 275 credits. Add 2 character sheets and 1 location image (15 credits) for 290 credits in total. Credits never expire, there is no subscription, and new accounts get 10 free credits. Failed steps are refunded automatically.
Start your cast
Add your characters, give each one a voice, and let the AI write a free season plan. Open the AI Drama Series Generator to begin.
FAQ
How do I keep an AI character's voice the same across videos?
Set a voice for the character once, from a description or a recording. In the drama generator, every episode then uses that voice on the Seedance models.
Is AI voice design for characters free?
Yes. Designing a voice from a description is free, with 3 tries per character.
What if I leave the voice description blank?
The tool makes a voice from the character's face text. If there is no face text, it uses the character's photo.
How long should an uploaded voice recording be?
2 to 15 seconds. Use a clean clip with one speaker, and only upload a voice you have consent to use.
How many characters can speak in one episode?
Up to 3 speaking characters per episode get their voice. A cast can have up to 4 characters.
If I change a voice, do old episodes update?
No. Finished episodes keep their old voice. Only episodes you make after the change use the new one.
Ready to try it?
Create your first AI video in minutes — no credit card required.
Start Creating Free →