Prompting Tips

The quality of your generated audio depends largely on how you describe what you want. This guide covers prompting techniques for every part of Audio Studio.

Story prompts

When starting a new project or chatting with Narra, give the AI enough context to build a rich audio experience.

Include these details:

• Setting. Where and when does the story take place?
• Characters. How many? What are their names and personalities?
• Mood and genre. Horror? Comedy? Sci-fi thriller? Cozy bedtime story?
• Audience. Is this for children, teens, or adults?
• Length preference. A short vignette or a longer narrative?

Good example:

"A tense noir mystery set in 1940s Chicago. Two characters: a weary private detective named Jack and a mysterious woman named Vera who walks into his office on a rainy night. Dark, suspenseful mood. For adult listeners."

Weak example:

"Write a detective story."

Sound effect prompts

For sound effects, describe the sound itself rather than the action or situation.

Be specific about:

• The sound source. What's making the noise?
• Environment. Indoor, outdoor, large space, small room?
• Intensity. Loud, subtle, distant, close-up?
• Texture and quality. Sharp, rumbling, metallic, soft, crisp?
• Duration hint. Brief impact or sustained ambient?

Good examples:

What to avoid:

• Don't describe what's happening in the story — describe what it sounds like.
• Don't use vague words like "cool sound" or "dramatic effect."
• Don't ask for multiple unrelated sounds in one prompt. Split them into separate segments.

Music prompts

Music prompts shape the genre, mood, and instrumentation of your background music.

Include these details:

• Genre. Jazz, orchestral, electronic, folk, ambient, etc.
• Mood. Tense, joyful, melancholic, triumphant, peaceful.
• Tempo. Slow, moderate, fast — or a specific BPM if you have one in mind.
• Instruments. Piano, strings, synth pads, acoustic guitar, percussion, etc.
• Vocal or instrumental. Specify if you want vocals (and in what style) or a purely instrumental piece.

Good examples:

"Gentle acoustic folk with fingerpicked guitar and soft violin, warm and nostalgic, slow tempo"

"Dark electronic ambient with deep bass pulses and high-pitched synth textures, unsettling and mysterious, very slow"

"Upbeat jazz trio — piano, upright bass, and brushed drums — lively and playful, moderate swing tempo"

Weak examples:

"Happy music"

"Something for a fight scene"

Voice design prompts

When designing a custom voice (rather than choosing a preset), describe the voice as if you were casting an actor.

Include these details:

• Age range. Young adult, middle-aged, elderly.
• Gender and vocal range. Male baritone, female alto, androgynous, etc.
• Accent or dialect. British RP, Southern American, Australian, etc.
• Vocal quality. Smooth, gravelly, breathy, warm, sharp, nasal.
• Personality and energy. Calm and authoritative, energetic and cheerful, cold and detached.

Good examples:

"A warm, deep male voice in his 50s with a slight Irish accent. Calm and reassuring, like a storyteller by a fireplace."

"A bright, energetic young woman in her 20s with an American accent. Confident and witty, with a hint of sarcasm."

"An elderly British woman with a soft, quavering voice. Kind and gentle, like a grandmother reading a bedtime story."

Weak examples:

"Male voice"

"Someone who sounds nice"

General prompting tips

These principles apply to all types of prompts in Audio Studio:

Be specific

Vague prompts get generic results. The more detail you provide, the closer the output will match your vision.

Use adjectives generously

Descriptive words give the AI concrete direction. "Whispered, urgent, breathless dialogue" paints a clearer picture than "quiet talking."

Say what you don't want

If there's something you want to avoid, mention it explicitly. "No percussion" or "not cartoonish" helps the AI avoid wrong directions.

Iterate and regenerate

Your first result might not be perfect. Tweak the prompt, regenerate, and compare. Small changes in wording can produce very different results.

Learn from what works

When you get a result you love, look at the prompt that created it. Use similar language and structure for future prompts.