How to Design a Voice from a Prompt

Voice Design creates a brand-new voice from a written description. You describe the voice you want — age, accent, mood, pace — and VocalLab generates variations for you to compare. Nothing is recorded and no audio sample is needed, which is what makes it different from voice cloning.

Watch the full walkthrough

Prefer to read? The written steps are below.

Step 1 — Describe the voice

Open Voice Design in Studio. The Voice Description box has two modes.

Basic — write it yourself

• Write the description in English, even if the voice will speak another language
• Between 30 and 1,000 characters
• Cover the things that actually shape a voice: age, gender, accent, pace, and tone

A description that works:

A middle-aged male voice with a clear British accent speaking at a steady pace and with a neutral tone.

If you'd rather start from something, four presets sit under the box — Support Agent, Narrator, Companion, and Meditation Instructor. Clicking one replaces whatever is in the box, so copy your own text somewhere safe first if you want to keep it.

Advanced — fill in the attributes

Switch to the Advanced tab and you get a field per attribute instead of a blank box: gender, age, emotion, tone, pitch, volume, speed, clarity, fluency, personality, texture, dialect, and audio quality.

Fill in only what matters to you — every field is optional, and anything you leave blank is left for the model to decide. The filled fields are combined into a structured description, shown underneath the form so you can see exactly what's being sent.

Advanced tends to give more predictable results than free text, because the attributes come out in the order the model responds to best. Basic gives you more room to describe something the fields don't cover. You can switch between them at any time.

Step 2 — Pick the spoken language and accent

The Language select controls what language the finished voice speaks. This is separate from your description, which stays in English either way.

Over 470 languages are available. The list starts with the ones that have a full voice catalog behind them — English, Spanish, German, French, Italian, Portuguese, Dutch, Polish, Russian, Chinese, Japanese, Korean, Arabic, Hebrew, and Hindi — followed by everything else alphabetically. The list is searchable, so type a few letters rather than scrolling.

Some languages also offer an Accent select, which appears next to Language when there's a choice to make:

Leaving Accent on its default lets the model pick. For every other language there's only one variant, so no Accent select appears.

Naming an accent here is more reliable than only asking for one in your description — though doing both does no harm.

Step 3 — Choose a script

Under Voice Script, pick one of two modes:

• Auto-generate script — VocalLab writes a short script that matches your description, in the language you selected. Easiest option, and the default.
• Write my own — supply your own text, between 50 and 600 characters. Aim for something that produces 5–45 seconds of audio.

Use Write my own when you want to hear the voice say something specific, like a line from the script you actually plan to narrate.

Step 4 — Generate and compare

Click Generate voice. Each run produces three variations — labelled Balanced, Warmer, and Brighter — so you can hear the same description interpreted three ways.

Play each one and click the card you want to keep.

Watch out: if you don't click a card, saving keeps the first variation (Balanced). Select the one you actually want before saving.

Not happy with any of them? Click Regenerate voice. This discards the current previews and costs points again.

Step 5 — Save it

Give the voice a name in Voice name (optional) — leave it blank and VocalLab names it for you — then click Save to My Voices.

Saved voices appear in My Voices under the Designed tab and can be used anywhere a voice picker appears, including the audiobook editor.

What it costs

Generating costs points. Saving is free.

Because each run creates three previews, a run costs roughly three times what generating that script once as speech would. Regenerating charges again — so it's worth spending an extra moment on your description rather than rolling the dice repeatedly.

The number of designed voices you can keep saved at once depends on your plan. See Plans Overview for your limit. On the free plan the limit is lifetime — deleting a designed voice does not free the slot back up.

Tips for better results

• Be specific about delivery, not just identity. "Warm, unhurried, slightly breathy" shapes the result far more than "nice female voice".
• Name an accent explicitly if it matters. Without one you'll get something neutral.
• Describe the use case. "Narrating a history documentary" gives the model more to work with than a list of adjectives.
• Don't stack contradictions. "Energetic but very calm" tends to average out into something flat.
• Two similar descriptions can sound quite different. If a run is close but not right, adjust one attribute at a time rather than rewriting the whole thing.

Troubleshooting

"Add a little more detail — at least 30 characters." Your description is too short. Add accent, pace, or tone.

"The description must be 1000 characters or fewer." Trim it. This is easiest to hit from the Advanced tab with every field filled in — the combined text counts toward the same limit, and the counter under the form shows where you are.

"Write a script for the voice to speak, or switch to auto-generate." You selected Write my own but left the script empty. Either write at least 50 characters or switch back to auto-generate.

"Your script is a bit short — use at least 50 characters for a good result." Short scripts don't give the model enough to work with. Lengthen it.

You've hit your design limit You're keeping the maximum number of designed voices for your plan. Delete one in My Voices, or upgrade. On the free plan the slot is permanent and can't be freed by deleting.

Out of points Generating previews costs points. Top up or upgrade, then try again.

Related

• How to Clone a Voice — use a real recording instead of a description
• Managing My Voices — organizing your saved voices
• How to Make an Audiobook — put your designed voice to work