ElevenLabs Voice Design: Create a Custom Voice From a Prompt

ElevenLabs Voice Design creates a brand-new synthetic voice from a written description, with no audio recordings and no real speaker to copy. Describe age, accent, tone and pacing, add preview text, generate, then save the option that fits. Each saved voice uses one custom voice slot, and plans range from 3 slots on Free to 2,200 on Business.

Last updated: October 2026. Plans and credit costs checked against ElevenLabs’ official pricing page.

What is ElevenLabs Voice Design?

ElevenLabs Voice Design is a tool that generates an original voice from a text prompt instead of from audio samples. It exists for the gaps in the Voice Library: when no existing voice has the right age, accent or character, Voice Design builds one to order. Saved voices then work in text to speech like any other.

“Speech synthesis is the artificial production of human speech.”

Source: Wikipedia, Speech synthesis

Voice Design takes that idea one step further: it synthesizes not only the speech but the speaker. Because the voice is invented rather than copied, it avoids the main ethical and legal question that hangs over cloning, which is whether the person whose voice is copied agreed to it. That makes it a practical choice for fictional characters, faceless channels and brand voices.

In one sentence: Voice Design invents a voice, while cloning copies one.

How is Voice Design different from cloning and the Voice Library?

Voice Design differs from voice cloning and the Voice Library in its input: a text prompt instead of recorded audio or a ready-made voice. Instant and Professional cloning copy a real speaker from samples, and the Voice Library offers voices other users have shared. Each route has different plan requirements and sharing rules.

Source: ElevenLabs pricing page, voice cloning docs and Voice Library docs (elevenlabs.io), checked October 2026.

Voice Design is the only route that needs no audio at all, but its voices stay private to the account.

For the cloning routes in detail, see ElevenLabs instant vs professional voice cloning; for ready-made voices and payouts, the ElevenLabs Voice Library guide.

Choose Voice Design when the voice needs to be new, cloning when it must match a real person.

How to create a custom voice from a prompt step by step

Creating a voice with ElevenLabs Voice Design takes six steps in the Voices section of the app, and the same feature is available through the API. The process is iterative: most usable voices come from two or three rounds of prompt tweaks rather than the first generation, so expect to refine before saving.

1. Go to Voices, then My Voices, then Add a new voice, and choose Voice Design.
2. Write a prompt describing the voice, starting with language and region in the first sentence.
3. Enter preview text for the voice to read, or click Auto-generate to create a sample script.
4. Adjust Loudness and Guidance scale if needed.
5. Click Generate and listen to each voice option it returns.
6. Select the best option and save it; it then appears in My Voices and uses one voice slot.

Write the preview text in the style the voice will actually perform, such as a product explainer or a story excerpt, so the audition reflects real use rather than a generic sentence.

Prompt, preview, generate, save: the loop is short, so iterate freely before committing a slot.

How do you write a voice design prompt?

A good voice design prompt describes how the voice sounds, not who the speaker is. The ElevenLabs docs recommend stating language and region first, then layering age, accent, pitch, timbre, pacing and audio quality in concrete terms. Precise, physical descriptions consistently beat abstract labels like professional or friendly on their own.

Here are the 6 attributes the official guidance recommends covering:

• Age: Phrases such as in their 20s or elderly man give the model a clear vocal age, which affects pitch, texture and energy more than almost any other single attribute.
• Accent: Specific regional accents work better than vague words, and the docs note that thick tends to work better than strong while foreign is too vague to steer anything.
• Pitch and gender sound: Describing the sound, such as a lower-pitched, husky female voice, gives the model acoustic detail instead of relying on an identity label alone.
• Tone and timbre: Physical texture words like warm, gravelly or airy describe the quality of the voice itself and shape how every line will sound in narration.
• Pacing: Instructions such as speaking slowly or staccato delivery set rhythm and speed, which matters for audiobooks, meditations and fast-cut social videos alike.
• Audio quality: Adding studio quality produces clean output, while poor audio quality deliberately roughens the sound, useful for phone calls or radio voices in fiction.

A prompt that covers these six attributes in two or three sentences is usually enough; stacking contradictory traits confuses the model more than it helps.

Describe the sound physically and specifically, language first, and the first generation gets much closer.

Voice design prompt examples that work

Voice design prompt examples are most useful when they show the jump from a vague request to a specific one. The stronger versions below follow the official structure: language and region first, then age, pitch, timbre, pacing and audio quality. Each targets a common creator use case.

Every stronger prompt names a language and region, an age, at least one texture word and a pace, which leaves far less to chance.

Copy the structure, not the words: language, age, sound, pace, quality.

What do Loudness and Guidance scale do?

Loudness and Guidance scale are the two sliders that adjust a Voice Design generation beyond the prompt. Loudness sets the output level of the preview, and Guidance scale, in the docs’ words, dictates how closely the prompt is adhered to. Higher guidance follows the prompt more strictly; lower guidance trades accuracy for creativity.

Here are the 5 controls on the Voice Design screen and when to use them:

• Prompt: The written description is the main lever, and it should lead with language and region, then add age, accent, timbre, pacing and audio quality details.
• Preview text: The script the voice reads during generation should match the intended use, since a calm story excerpt and an energetic ad read reveal very different qualities.
• Auto-generate: This button writes preview text automatically and includes audio tags, a quick option when exploring many prompts before writing custom scripts.
• Loudness: This slider raises or lowers the output level of generated previews, useful for keeping auditions comparable when switching between quiet and loud character voices.
• Guidance scale: Higher settings keep the voice close to the prompt, while lower settings allow unexpected but sometimes better results when the description is loose or experimental.

If outputs ignore part of the prompt, raise Guidance scale; if they all sound too similar, lower it.

Use Guidance scale as the dial between precision and surprise.

How many custom voices can each plan save?

Each ElevenLabs plan includes a fixed number of custom voice slots, and the docs state that a saved designed voice takes up one of them. Free includes 3 slots, Starter 10, Creator 30, Pro 160, Scale 660 and Business 2,200. Generating previews does not use a slot; only saving does.

Source: ElevenLabs pricing page (elevenlabs.io/pricing), checked October 2026.

Slots rise steeply with plan tier, so slot count only becomes a constraint for heavy users on Free or Starter.

Slots are shared with cloned voices, so a creator on Starter with a few instant clones has fewer slots left for designed characters. Deleting unused voices frees slots for new designs.

Save sparingly on Free and Starter; slot limits stop mattering from Creator upward.

What does Voice Design cost in credits?

Voice Design costs credits based on the preview text, not the prompt or the number of options returned. According to the ElevenLabs docs, each generation is charged once, based on the characters in the preview text, even though several voice options come back. Short preview scripts therefore keep experimentation cheap on any plan.

In practice that means the prompt can be as detailed as needed without raising the cost, while the preview text should be just long enough to judge the voice: a few sentences that include the kinds of words and emotions the final script will contain. After saving, the voice is billed like any other in text to speech, where ElevenLabs lists about one credit per character on Multilingual v2. The ElevenLabs text to speech settings guide covers tuning the voice once it is in use.

Detailed prompts are free; long previews are what cost credits.

Using a designed voice in narration, agents and video

A designed voice is used exactly like any other voice in an ElevenLabs account once it is saved to My Voices. The docs state designed voices work with Eleven v4 and earlier models, including Eleven v3, so they can narrate scripts, power conversational agents and voice faceless videos without extra setup.

Here are the 5 most practical uses for a designed voice:

• Faceless YouTube channels: A distinctive narrator built from a prompt gives a channel a consistent, recognizable sound without the creator recording or cloning their own voice for every upload.
• Fiction and audiobook characters: Separate designed voices for each character, varied by age, accent and timbre, make multi-voice stories easier to follow than a single narrator switching styles.
• Conversational agents: A brand-specific voice for a support or booking agent avoids sounding like other businesses that use the same popular library voices on their phone lines.
• Ads and explainers: Marketers can design several voices aimed at different audiences and test which one performs best, without booking and paying multiple voice actors for auditions.
• Games and prototypes: Non-player characters, narrators and placeholder dialogue can be voiced early in development, then kept or replaced once the game’s final audio direction is settled.

For agents, see how to build an ElevenLabs voice agent; for channels, how to create faceless YouTube videos with AI.

Once saved, a designed voice is a first-class voice everywhere in ElevenLabs.

Common Voice Design mistakes to avoid

Common Voice Design mistakes come from treating the prompt like a casting call instead of an acoustic description. The ElevenLabs docs flag several specific pitfalls, from vague accent words to effects terms the model does not handle well. Avoiding them saves generations, credits and the frustration of near-miss voices.

Here are the 5 mistakes that waste the most generations:

• Using effects words: The docs advise against FX terms such as reverb or echo, since Voice Design shapes the voice itself rather than adding post-production effects to it.
• Confusing accent with intonation: Writing accent when describing melody or emphasis misleads the model, so describe rising, flat or sing-song delivery directly instead of calling it an accent.
• Vague origin words: Terms like foreign give no usable signal, while naming a specific country or region gives the model a concrete accent target to aim for.
• Burying the language: Putting language and region late in a long prompt weakens their effect, and the docs recommend stating them in the very first sentence.
• Generic preview text: A bland test sentence hides how the voice handles emotion and pacing, so preview text should resemble the real script the voice will perform.

Most bad results trace back to vague or effects-based wording, not to the model.

Commercial use and sharing limits for designed voices

Designed voices can be used commercially on paid plans, since the Commercial License is listed from Starter upward and the Terms limit free use to non-commercial purposes. They cannot, however, be published to the public Voice Library: only Professional Voice Clones can be shared there, so designed voices stay private to the account that made them.

Here are the 5 rules that govern designed voices:

• Free plan restrictions: Voices designed and used on Free fall under the non-commercial limit in the Terms, so monetized videos and client projects should run on Starter or higher.
• Commercial license from Starter: The pricing page lists the Commercial License under Starter, the cheapest plan where a designed voice can narrate monetized or client-facing content.
• No public sharing: Only Professional Voice Clones can be published to the Voice Library, so designed voices cannot earn library payouts and remain usable only inside the creating account.
• Output ownership: The ElevenLabs Terms state users retain all rights in and to their output, which covers audio generated with a designed voice on an eligible plan.
• Slot accounting: Every saved designed voice occupies one custom voice slot, shared with cloned voices, so plan limits cap how many characters an account can keep.

Paid plan for commercial work, private use only: those are the two rules that matter most.

Bottom line: is ElevenLabs Voice Design worth using?

ElevenLabs Voice Design is worth using whenever a project needs a voice that does not exist yet and should not be copied from a real person. It needs no recordings, charges only for preview text, and saved voices work across narration, agents and video. Its limits are slot counts on lower plans and no public sharing.

“Voice Design lets you create a unique voice from a text prompt.”

Source: ElevenLabs Docs

“Voice Design is the simplest route to a voice nobody else on the platform is using.”

Source: GrowthStackKit, ElevenLabs Review

For how Voice Design fits with cloning, dubbing and the rest of the platform, see the complete ElevenLabs review. A good first exercise is designing one narrator with a short, specific prompt and saving it only once it reads the real script well.

Precise prompts, short previews and a paid plan for commercial use make Voice Design a strong, low-risk tool.

Frequently asked questions

The 12 most-asked questions about ElevenLabs Voice Design.

What is ElevenLabs Voice Design?

ElevenLabs Voice Design is a tool that generates an original synthetic voice from a written description instead of recorded audio. It fills gaps when the Voice Library has no suitable voice. A creator describes age, accent, tone and pacing, generates options, and saves the best one to use in text to speech.

How do I use ElevenLabs Voice Design?

Go to Voices, then My Voices, then Add a new voice, and choose Voice Design. Write a prompt describing the voice, add or auto-generate preview text, adjust Loudness and Guidance scale, then click Generate. Listen to the options and save the best one, which uses one voice slot.

How much does ElevenLabs Voice Design cost?

According to the ElevenLabs docs, each Voice Design generation is charged once based on the characters in the preview text, not the prompt length or the number of options returned. Saved voices are then billed like other voices in text to speech, which the pricing page lists at about one credit per character on Multilingual v2.

How many voices can I create with Voice Design?

Saved designed voices count against custom voice slots: 3 on Free, 10 on Starter, 30 on Creator, 160 on Pro, 660 on Scale and 2,200 on Business. Generating previews does not use a slot. Slots are shared with cloned voices, so deleting unused voices frees space for new designs.

Is Voice Design the same as voice cloning?

No. Voice cloning copies a real speaker from audio samples, with Instant Voice Cloning from Starter and Professional Voice Cloning from Creator. Voice Design invents a new voice from a text description with no recordings, which avoids the consent questions that come with copying a real person’s voice.

Can I share a designed voice in the Voice Library?

No. Only Professional Voice Clones can be published publicly to the ElevenLabs Voice Library, after review. Voices made with Voice Design or Instant Voice Cloning cannot be shared there, so they stay private to the account that created them and cannot earn Voice Library payouts.

Can I use a designed voice commercially?

Yes, on a paid plan. The Commercial License is listed from Starter upward, while the Terms restrict free use to non-commercial purposes. The Terms also state that users retain all rights in their output, so audio produced with a designed voice on an eligible plan can be used in monetized content.

What should a Voice Design prompt include?

Start with language and region, then describe age, accent, pitch, timbre, pacing and audio quality. Concrete phrases such as in their 20s, thick Scottish accent, warm and gravelly, speaking slowly and studio quality work well. Avoid effects words like reverb or echo, and avoid vague terms like foreign.

What does Guidance scale do in Voice Design?

Guidance scale controls how closely the generated voice follows the prompt. Higher values stick to the description more strictly, while lower values allow more creative results at the cost of prompt accuracy. Raising it helps when outputs ignore part of the prompt; lowering it helps when every option sounds too similar.

Which models work with designed voices?

The ElevenLabs docs state that designed voices work with Eleven v4 and earlier models, including Eleven v3. That means a saved designed voice can be used for expressive narration, conversational agents and other text to speech work in the same way as library or cloned voices.

Can I change a designed voice after saving it?

The docs do not describe editing a saved designed voice. The practical approach is to refine the prompt and preview text before saving, then generate and save a new version if the voice needs to change, deleting the old one afterward to free its custom voice slot.

Is Voice Design available through the API?

Yes. The ElevenLabs docs state that Voice Design is available through the API as well as in the app, so developers can generate and save designed voices programmatically. API requests use the same account API key and draw on the same credits and custom voice slots as the web app.