How to write AI music prompts
Give the music a purpose, an opening, one clear change and an ending. Explore a real Aqua request and its result, then assemble a brief you can edit before taking it into the studio.
The workbench runs in your browser without an account. It does not create audio, submit jobs or use generation credits.
Four parts of a useful musical brief
Purpose: one scene and one feeling
“A hopeful instrumental for a quiet reunion” gives a clearer starting point than a string of unrelated styles. Say whether you want singing or an instrumental. Use your own story and describe musical qualities rather than asking for a copy of a named artist.
Opening: a small set of instruments
Name the first sound and how much space it has. For example, “Begin with a simple felt-piano motif and solo cello.” Save the full ensemble for later if you want the piece to grow.
Change: one development that matters
Ask for one arrival, contrast or lift: strings answering the piano, a chorus opening up, or percussion entering. A long sequence of competing events can be difficult to fit into a short render.
Ending: a deliberate finish
Return to the opening instrument, settle the energy or ask for a standalone ending. These are creative directions. A generated ending is not automatically a seamless loop, a frame-accurate film cue or a clean advertising edit point.
In Aqua, put the overall direction in Describe your song. Use Instruments & texture, mood and Styles and sounds to avoid for a few supporting details. Put sung words in the separate reviewed-lyrics field; arrangement instructions are not a lyric sheet.
First Light, Far Horizon
This is a 35-second excerpt, from 40–75 seconds of an existing 120-second Aqua instrumental. The original request below was recovered from its saved song and render records. The completed audio was mastered after generation.
AI-generated instrumental. One prior result, not a live render or a controlled comparison. The excerpt does not demonstrate the requested opening or final button, and the written request is not proof that every instrument or instruction was followed.
The exact submitted main prompt
Original emotional orchestral advertising score. Begin with a memorable felt-piano motif in spacious silence. Introduce solo cello and chamber strings, then widen the scene with harp and warm French horns. Build gradually into soaring strings and restrained cinematic percussion before resolving gently with a clean final button for a logo. Use premium acoustic detail, clear edit points, natural dynamics, and one coherent original motif.
- Instruments requested
- Felt piano, chamber strings, solo cello, French horns, harp, soft timpani
- Mood requested
- Tender, reflective, reassuring, hopeful, quietly triumphant
- Sounds to avoid
- Vocals, choir, trailer braams, heavy drums, horror dissonance, abrupt edits, existing themes, artist imitation
- Original settings
- Instrumental · ACE-Step · 120 seconds · 72 BPM · D major
What the request teaches
It introduces a motif, adds instruments in stages and asks the ending to return to a coherent idea. The instrumentation and mood fields support that same direction.
What it cannot establish
One result cannot show which instruction caused a sound. Reusing the words does not guarantee the same performance, exact timing or instrumentation. For your first short render, simplify the request.
Open this instrumental starter
After signup and email verification where required, the studio opens with an editable instrumental starter. Review the current length and displayed cost. Nothing is generated until you choose to submit.
Build your own four-part prompt
Keep each part brief. This workbench joins your directions into editable text; it does not use a model or judge whether the music will work. The prefilled directions are an illustrative starting point, not the recorded request above. Your entries stay on this page and are not saved, placed in a URL or sent to the studio.
Name a scene and say whether it is singing or instrumental.
Start with a few instruments and a clear amount of space.
Choose one development or lift that fits your intended length.
Give the piece a direction for its final moments.
You can edit this text directly. Assembling again replaces it with the four fields above.
Review the words, then copy them into Describe your song. Your custom brief is not transferred automatically.
Want a singer? Use reviewed lyrics and an original generated singer, or set up a voice you are authorised to use. Keep vocal-delivery notes separate from the sung words. An original generated singer does not require uploading another person’s voice.
Two prompts with different jobs
These original teaching examples have not been rendered for this guide. They illustrate the four-part method; the audio above came from its separately recorded request.
A reading track that stays out of the way
A quiet instrumental for focused reading, calm and steady. Begin with soft electric piano and warm bass at a gentle pace. Introduce brushed percussion without a large chorus or sudden lift. Keep a small recurring motif and end softly. No vocals, sharp drops or busy lead solos.
Why these directions fit together: the small instrument list, restrained development and soft ending all support attention rather than a dramatic payoff. Check the actual result for intrusive changes; the prompt alone cannot guarantee unobtrusive music.
A vocal arrangement that leaves room for an apology
An original acoustic song about apologising without asking for immediate forgiveness. Begin with close fingerpicked guitar and a restrained original AI singer. Add warm piano and light brushed drums for one fuller chorus, keeping the lead words clear. Return to guitar for a quiet ending. Use the supplied reviewed lyrics; no extra verses or dense backing choir.
What still needs your input: write or choose lyrics you have the right to use, then place them in Use my reviewed lyrics. This is arrangement direction, not a completed lyric or an invitation to imitate a real singer. Hear Aqua’s existing vocal demonstration.
Change what you actually hear
| If the result… | Try one focused revision |
|---|---|
| Feels too busy | Remove secondary instruments or ask for fewer arrangement changes. Keep the most important melody or voice. |
| Never develops | Name one arrival: a second instrument enters, percussion lifts a chorus or the harmony opens up. |
| Rushes the words | Shorten the lyric or give it more time within the lengths available to your account. Avoid adding more instructions to an already crowded brief. |
| Misses the finish | Simplify the requested structure and ask for one clear ending. Check the final audio rather than assuming the instruction was followed. |
Compare each render with its saved brief and note what changed. Random generation can change more than the field you edited, so a different result alone does not prove that one phrase caused the improvement. Listen for altered or missing lyrics, unwanted vocals, artefacts and abrupt transitions before sharing.
A free workbench, a limited render queue
The text workbench is free to use. Audio generation uses Aqua’s public beta: email verification activates 30 credits, a standard song costs 10 credits, and the current public song-length cap is 120 seconds. No card is required and payments are disabled. Grants and admission are capacity limited.
Aqua’s local worker processes one song at a time. Expect minutes, with more waiting when the queue is busy; failed jobs release reserved credits. Songs stay private unless you choose to share a public link. A successful prompt is not a guarantee of exact lyrics, instrument choices, musical quality or rights for every use.
Music prompt questions
Should I paste lyrics into the music description?
Use Describe your song for musical direction and the separate Use my reviewed lyrics field for words to be sung. Choose an original generated singer or an authorised saved voice. For instrumentals, choose Instrumental · no vocals and No lyrics.
Will the recorded prompt recreate First Light exactly?
No. It is a saved original request beside one existing result. Models, settings and random generation affect the output; matching the words does not promise the same recording, instrumentation or edit points.
Does the workbench send or save my writing?
The workbench assembles text in this page. It does not send the field contents to a server, put them in a URL, save them in browser storage or submit a generation. Copy your brief if you want to keep it; it may be lost when you leave. The separate recorded-starter link carries only a fixed public example ID.
Is generated music automatically royalty-free?
No automatic rights guarantee is made here. Use lyrics, voices and reference audio you own or have permission to use, review the result and check the rights needed for your intended project. Read Aqua’s copyright policy.
Where do I learn the account and generation steps?
The first-song guide covers signup, verification, the studio controls, rendering and downloads. This guide focuses on writing and refining the musical brief.