How to Create AI Character Voices for Stories
Learn to create AI character voices for stories: design a voice from a description, keep it consistent across scenes, direct each line and follow voice rules.
To create an AI character voice, pick one of 30 prebuilt voices, design one from a text description, or clone one from a recording you have permission to use. Set the voice once, write a one-line brief, and direct each line with a short style and a few tags. OndaLabs supports all three routes.
Three ways to get a character voice
Each route trades effort against control. The table shows what each one takes and what it costs in OndaLabs.
| Route | What you provide | OndaLabs cost or limit | Industry example | |---|---|---|---| | Prebuilt voice | Pick one of 30 | Included on every plan | ElevenLabs' library holds more than 10,000 community voices | | Design from a description | A text description | 1,500 credits; custom voice slots on Creator (3), Pro (10) and Business (30) | ElevenLabs takes a description of 20 to 1,000 characters and returns 3 previews | | Clone from a recording | A recording you have permission to use | 1,500 credits; the same custom voice slots | ElevenLabs' instant cloning works from samples under two minutes |
Industry examples come from ElevenLabs' voices overview, as fetched on 2026-10-17. OndaLabs figures come from its published plans and costs.
Design or clone: choose by where the voice comes from
Design a voice when the character does not exist yet. You describe the voice in words, generate it, and keep the version that fits. The description matters most, so spell out the attributes you care about: age, gender, accent and tone. Vendors differ on how many options they return and how long a description can be. ElevenLabs accepts 20 to 1,000 characters and returns three previews, but check what OndaLabs gives you when you design a voice.
Clone a voice only when you have a recording of that voice and permission to use it. ElevenLabs' instant cloning works from samples under two minutes, and its professional cloning needs a Creator plan or above. Cloning suits a host or narrator who records their own voice. It does not suit a character who should sound like someone else. In OndaLabs, both routes cost 1,500 credits and use a custom voice slot.
Once the voice is set, the next job is the tone of each scene. See emotional text to speech for directing lines, and AI voices for games if the character is for a game.
Write a one-line brief for each character
A short brief keeps every decision about the character in one place. Write one line per character with these parts:
- Age and gender: choose these through the voice. Google's guidance is to keep them out of the per-line style.
- Accent: describe it in the voice design. Results depend on the description, and this post makes no promise about any specific accent.
- Tone and personality: one or two adjectives, such as "warm, quick and a little nervous".
- Role in the story: what the character wants in this scene.
- What to keep out of line directions: names and permanent traits belong in the brief, not in each line.
A brief might read: "Older tavern keeper, low and gravelly, suspicious of strangers." Nothing in that line has to change from scene to scene, which is the point.
Keep a character consistent across scenes
Consistency comes from how you work, not from one setting. Google's guidance is to "Design the persona once with Voice design, then keep style short or empty". It also advises against extra meta-instructions such as "maintain identical timbre", because they increase drift.
In practice, keep the same voice selected for the character, keep line directions short, and generate a fixed test phrase before each session so you can compare the output with earlier takes. No tool can promise an identical result every time, so treat this as a method for getting close, not a guarantee.
Direct each line with style and tags
Direction works best when the character's baseline is in the voice and the line-level direction handles the moment. A fantasy scene in the product shows the idea, with a witch and a knight using <whispers>, <cackle> and <sigh> at key points. The game example pairs a gravelly voice with the direction "gruff, suspicious tavern keeper, low and gravelly", plus <tsk> and <chuckle>.
The product tip is still the safer default: "Keep directions short: one or two words per line." Use longer directions sparingly, and reserve tags for the moments that need them.
Cast two characters in one scene
The Dialogue studio handles a scene with two characters. Each speaker has a name and a voice. Names use letters and numbers only, up to 30 characters, and the two names must differ. A reaction inside a line, such as |oh hmm|, lets one character respond to the other. The studio's job interview example pairs Grace, with the mature-sounding Gacrux voice, and Ben, with the clear Iapetus voice. Those are product examples, not claims that these voices suit your character.
Legal and ethical checks
Clone only voices you have permission to use. Do not assume you can imitate real actors or famous characters. Tennessee's ELVIS Act, signed on 2024-03-21 and effective 2024-07-01, protects a person's voice, including a simulation of it (Nelson Mullins). The federal NO FAKES Act is still a bill, not law, as of 2026-10-17. It was reintroduced on 2026-05-20 as S.4591 and H.R.8915 (Manatt). For the full picture, see AI voice cloning laws in 2026. This is general information, not legal advice.
Make AI character voices in OndaLabs
Start with the free plan, which includes dialogue on Onda Lite and all 30 prebuilt voices. Test your character on a short scene before you commit to a voice. Use the emotional peak of the scene as your test line, because it shows the most about how the voice handles feeling.
Keep the one-line brief for each character in your project notes, next to the voice's name. Anyone who joins the project later can then match the character without guessing what the voice was meant to be. When you are ready to design or clone a recurring character, use the custom voices area. Each design or clone costs 1,500 credits. You can compare plans on the pricing page, or sign up to start.
Frequently asked questions
How do I create an AI voice for a fictional character?
Start with a one-line brief that covers age, tone, accent and role. Choose a prebuilt voice that fits, or design a voice from a text description, and keep that voice for every line. Direct each line with a short style and a few tags.
Can I design a voice from a text description?
Yes. In OndaLabs, designing a voice costs 1,500 credits and uses one of your custom voice slots, which are 3 on Creator, 10 on Pro and 30 on Business. ElevenLabs uses a description of 20 to 1,000 characters. Results depend on the wording, so generate a test phrase and adjust the description if needed.
How do I keep a character's voice consistent across scenes?
Use the same voice every time, keep per-line directions short, and leave age, gender and accent out of the directions. Generate a fixed test phrase before each session and compare it with earlier output. No tool can promise an identical result every time.
Do I need permission to make a voice that sounds like a famous character?
Do not assume you can. Avoid imitating real actors or famous characters. State laws such as Tennessee's ELVIS Act protect a person's voice, including simulations, and the federal NO FAKES Act is still a bill. Check the law where you publish, and get legal advice for a specific project.
Can AI voices do accents, ages and creatures?
You can describe an accent, age or creature quality in the voice brief, but results depend on the voice and the wording. OndaLabs does not promise specific accents. Listen to a test phrase before you commit to a cast.