Creative tools

Designing better custom voices

A custom voice is shaped by two vital inputs: the voice description and the sample text. The description explains what kind of speaker to create. The sample text gives the model a line to perform in that identity. Both influence the voice you save.

The description and sample text work together

The description sets the foundation for the voice: tone, personality, emotion, pace, and context. The sample text gives the model a reference for delivery, phrasing, timing, and emotional nuance.

The model interprets both inputs together. It does not treat the description as a checklist of isolated traits. It imagines the speaker in context, then performs the sample line through that personality.

This means the same description can produce different results with different sample text. Iterate on both: revise the description, try alternate sample lines, and save the version where the voice identity and the performed line feel aligned.

Describe the speaker, not just the sound

Good voice prompts combine vocal quality, personality, delivery, and context. Think like you are casting an actor for a role.

The speaker has a warm, lightly gravelly voice with dry comic timing, like a beloved stage actor delivering a clever aside with total confidence.

Useful details to include

Tone: warm, stern, playful, melancholy, skeptical, gentle.
Pace: measured, quick, clipped, unhurried, breathless.
Texture: clear, smoky, nasal, gravelly, crisp, airy.
Performance style: naturalistic, theatrical, intimate, announcer-like, documentary.
Context: audition reader, narrator, villain, parent, news anchor, musical theatre lead.

Choose sample text that fits the voice

The sample text should sound like something this character or narrator would actually say. It is not just a throwaway preview: it helps shape the saved voice by showing how the speaker handles real language, rhythm, punctuation, and emotion.

It does not need to be from your script, but it should test the delivery you care about. If the voice needs to be sarcastic, give it a sarcastic line. If it needs to be tender, give it a tender line.

Keep it readable

Use complete sentences and normal punctuation. Avoid emojis, Markdown, HTML, or long lists of disconnected adjectives.

The speaker has a precise, elegant British voice, calm under pressure, with a subtle edge of impatience and a controlled, aristocratic delivery.

Prompt patterns

Narrator

The speaker has a calm documentary narrator voice, observant and intelligent, with a steady pace, clean articulation, and quiet authority.

Musical character

The speaker has a bright musical theatre voice, expressive and emotionally open, with crisp diction, youthful optimism, and a strong sense of rhythm.

Antagonist

The speaker has a smooth, controlled voice with restrained menace, soft-spoken confidence, and a deliberate pace that makes every word feel chosen.