> ## Documentation Index > Fetch the complete documentation index at: https://docs.breezeblue.ai/llms.txt > Use this file to discover all available pages before exploring further. # Voice Design prompting > Describe a reusable voice, choose a preview script, and evaluate the result. Describe the voice you want in `voice_description`. A short prompt such as “a young woman with a light French accent” is enough to start; add details for a more specific result. ## Quick example Send the voice description and optional preview script as separate fields to `POST /v1/voice-previews/design`: ```json theme={null} { "voice_description": "A young woman with a light French accent and a warm, clear voice. Speak at an easy, conversational pace.", "text": "Welcome. Let me show you around and help you get settled.", "guidance_scale": 3 } ``` `voice_description` is required; `text` is optional. Each accepts 3–500 characters when supplied. ## Writing the prompt Choose any of these optional dimensions; more detail does not guarantee a better voice. * **Role and situation:** Who is speaking, and where? For example, “a late-night radio host speaking to someone who cannot sleep.” * **Base voice:** Use everyday terms for age, gender, pitch, accent, or synthetic character: “a middle-aged man with a deep, warm voice.” * **Vocal texture:** Pick one or two audible traits, such as “soft and breathy” or “slightly rough.” * **Pace and delivery:** Describe habitual rhythm and articulation: “slow and deliberate” or “clear pronunciation with gentle pauses.” Use TTS `instructions` for individual performances. * **Emotion and script fit:** Describe a natural emotional center, such as “tired but professional,” without prescribing the emotion of every future line. For optional preview `text`, use words the speaker would naturally say to hear how the voice fits its role. ## Examples ### Meditation guide **Voice description** ```text theme={null} A calm female meditation guide with a soft, breathy voice. Speak slowly and smoothly, with gentle pauses and a warm, reassuring tone that feels peaceful, intimate, and safe. ``` **Preview text** ```text theme={null} Let your shoulders drop away from your ears, releasing any tension you might be holding. Breathe in deeply through your nose, feeling the air expand your chest, and then slowly exhale, letting everything soften into this present moment. ``` ### Elegant villain **Voice description** ```text theme={null} A middle-aged man with a deep, smooth voice, playing an elegant villain. Speak slowly and clearly with controlled menace; keep the tone calm, intimate, and faintly amused, never loud. ``` **Preview text** ```text theme={null} Did you truly believe you could dismantle my life's work without consequence? I admire the audacity, really, I do. But now that the doors are locked, let us discuss exactly how much your ambition is going to cost you. ``` ### Older chronicler **Voice description** ```text theme={null} An older man with a deep, slightly rough and weathered voice. Speak slowly with measured pauses and restrained emotion, giving narration, warnings, and dark memories a quiet sense of weight. ``` **Preview text** ```text theme={null} People think ruin announces itself with thunder. It does not. It sits beside you quietly, waits for compromise, and learns your name before you notice the door is locked. ``` ### Polished digital assistant **Voice description** ```text theme={null} A friendly digital assistant with a bright, clear voice and a subtle synthetic quality. Speak at a crisp, steady pace with precise pronunciation, sounding warm, helpful, and consistently composed. ``` **Preview text** ```text theme={null} Thank you for waiting. I have located your order details and will now summarize the current status, estimated delivery window, and available support options. ``` ## Guidance `guidance_scale` controls how closely the output follows your voice description and preview script. Higher values strengthen guidance; lower values relax it. Try **2–4** as a starting range and audition the result. The accepted range is **1–10**. Omitting the value lets BreezeBlue sample the guidance strength automatically; the suggested range is not the API default. ## Draft a voice description with your own LLM Ask the LLM in your agent harness to turn the character brief into a voice description and a separate preview script. No BreezeBlue call is needed to draft them. ```text theme={null} Turn this brief into JSON with "voice_description" and "text". Write both in English, with each field between 3 and 500 characters. Describe a reusable voice: role, pitch, accent, one or two audible texture traits, and habitual rhythm. Keep temporary scene emotions out of the identity. Make "text" a representative line spoken by this character, not directions. Do not promise an exact match to a real person's voice. Brief: An older museum guide with a low, slightly rough voice, General American pronunciation, measured pacing, and quiet curiosity. ``` Review the description for audible, compatible traits. Keep the description and preview script in the intended speech language; when setting `language_code`, choose a language supported by the selected model. Pass `voice_description` to CLI `--description` and the preview script to `--text`: ```bash theme={null} breeze voice design \ --description "An older museum guide with a low, slightly rough voice. Speak General American English with measured pacing and quiet curiosity." \ --text "Look closely at the worn edge. It tells us how this tool was held." \ --language en --guidance-scale 3 --preview-count 2 --agent ``` Generating previews consumes credits. Audition accent, texture, clarity, and suitability across representative lines. Save the chosen preview to obtain a reusable voice ID, then use TTS `instructions` for the emotion and delivery of each passage. ## Next steps Generate previews, audition them, and save a reusable voice. Generate encoded previews and compare multiple candidates. Direct the situation, intent, and delivery of each passage. Generate speech with HTTP, SDKs, or asynchronous jobs.