Music Prompts
Describe what you hear, not what you want.
The best music prompts read like liner notes, not feature requests. Genre is the container. Mood is the content. Instrumentation is the texture. Structure is the architecture.
The Template
| Element | What It Controls | Example |
|---|---|---|
| Genre | Sonic foundation | "Lo-fi hip hop" |
| Mood | Emotional register | "Nostalgic, bittersweet" |
| Instrumentation | Sound palette | "Rhodes piano, vinyl crackle, soft kick" |
| Structure | Song architecture | "Intro 8 bars, verse, chorus, bridge, outro fade" |
| Tempo | Energy and pace | "82 BPM, relaxed swing feel" |
| Vocal | Voice direction | "Female vocal, breathy, reverb-heavy, no vibrato" |
| Reference | Sonic anchor | "In the style of Nujabes meets Boards of Canada" |
Techniques
Genre Blending
Novel sounds live at intersections.
"Afrobeat rhythm section meets Scandinavian ambient textures. Talking drum pattern under shimmering pad synths. 110 BPM."
Mood Layering
Stack emotional registers for depth.
"Start melancholic — solo piano, sparse. Build hope — add strings at verse 2. Arrive triumphant — full band, brass hits on chorus. The journey is grief to gratitude."
Structural Direction
Tell the model where the song goes, not just what it sounds like.
"32-bar intro builds tension with rising synth pad. Drop at 0:33 — full beat, bass, and vocal chop hook. Bridge strips to just vocal and piano. Final chorus doubles the energy."
Lyric Prompting
Separate lyric generation from music generation for better results.
"Write lyrics about leaving a hometown. Tone: grateful, not sad. Use specific sensory details — the smell of salt air, a screen door sound, a specific street name. No cliches about 'finding yourself.'"
The Prompt
A complete song generation prompt for Suno or Udio. Uses every template element: genre, mood, instrumentation, structure, tempo, vocal direction, and reference.
GENRE: Indie folk meets electronic — acoustic guitar foundation with
subtle synth textures and programmed beats. Think Bon Iver's "22, A
Million" crossed with The Tallest Man on Earth's rawness.
MOOD: Bittersweet nostalgia. The feeling of driving through your
hometown 10 years after leaving. Not sad — grateful with an ache.
INSTRUMENTATION:
- Foundation: fingerpicked acoustic guitar, warm and close-mic'd
- Texture: ambient pad synth (low, barely there), vinyl crackle
- Rhythm: soft programmed kick (not four-on-the-floor), brushed snare
- Accent: distant piano chords on chorus only, pedal steel ghost notes
- Absence: NO full drum kit, NO electric guitar distortion, NO brass
STRUCTURE:
[Intro — 8 bars] Fingerpicked guitar alone, establish the key
[Verse 1 — 16 bars] Vocal enters, intimate and close. Add synth pad.
[Pre-chorus — 8 bars] Build with layered vocals, rising dynamic
[Chorus — 16 bars] Full arrangement, piano enters, vocal opens up
[Verse 2 — 16 bars] Strip back to guitar + voice, new lyric detail
[Bridge — 8 bars] Just vocal and piano. Most vulnerable moment.
[Final Chorus — 16 bars] Everything returns, add harmony vocals
[Outro — 8 bars] Instruments fade one by one. Guitar last. No fade-out
— let the final note ring and decay naturally.
TEMPO: 94 BPM. Relaxed but not sluggish. Slight swing feel.
VOCAL DIRECTION: Male vocal, mid-range baritone. Close and breathy on
verses. Opens to chest voice on chorus. No falsetto. No vocal runs.
Think Gregory Alan Isakov's restraint.
LYRIC THEME: Write about a specific place — not "the old town" but the
name of a street, the sound of a screen door, the way light hits a
particular wall at 4pm. Sensory details over sentiment. Grateful tone,
not sad. No cliches about "finding yourself" or "coming home."
Paste into Suno or Udio. Use Claude to generate lyrics first, then feed both into the music tool.
Tools
| Tool | Strength | Link |
|---|---|---|
| Suno | Full song generation, vocals, lyrics | suno.com |
| Udio | Audio fidelity, genre accuracy | udio.com |
| Stable Audio | Open weights, fine-tuning | stability.ai/stable-audio |
| ElevenLabs | Voice cloning, speech synthesis | elevenlabs.io |
| AudioCraft | Meta's open-source audio generation | github.com/facebookresearch/audiocraft |
Context
- Audio — Audio tool capabilities and comparison
- Creativity Prompts — Ideation techniques
- Prompts — First principles across all modalities
Every song is a feeling with structure.
Questions
When you describe music in words, what gets lost in translation — and is that loss generative?
- What separates a prompt that produces a generic beat from one that produces a vibe?
- If AI can generate any genre, does genre still mean anything?
- At what point does "in the style of" stop being a reference and start being a replacement?