A Guide to Music Generation Prompting with Elevenlabs

Good songs tell stories, create feelings, and make people move. They don't explain themselves, apologize for existing, or try too hard to be clever.

Recent Project Examples

Quick Start

Basic API Call

curl -X POST <https://api.elevenlabs.io/v1/music/detailed> \\
     -H "Content-Type: application/json" \\
     -H "xi-api-key: YOUR_KEY" \\
     -d @your_composition.json \\
     --output song.mp3

Minimum Viable Composition

{
  "composition_plan": {
    "positive_global_styles": ["electronic pop", "125 BPM"],
    "negative_global_styles": ["slow", "acoustic"],
    "sections": [{
      "section_name": "Full Song",
      "positive_local_styles": ["energetic"],
      "negative_local_styles": ["boring"],
      "duration_ms": 60000,
      "lines": ["Your lyrics here"]
    }]
  }
}

JSON Structure Deep Dive

Global Styles (Apply to Entire Song)

"positive_global_styles": [
  "genre",           // Primary: "electronic pop", "dark techno", "hyperpop"
  "BPM",            // Critical: "125 BPM", "140 BPM", "90 BPM"
  "vocal_style",    // Character: "sultry", "aggressive", "theatrical"
  "production",     // Sound: "bass-heavy", "synth-driven", "minimal"
  "mood"           // Feeling: "chaotic", "mysterious", "playful"
]

Section Anatomy

{
  "section_name": "Chorus",        // Standard: Intro, Verse, Pre-Chorus, Chorus, Bridge, Outro
  "duration_ms": 24000,           // 3000-120000ms per section
  "positive_local_styles": [],     // Section-specific additions
  "negative_local_styles": [],     // Section-specific exclusions
  "lines": []                      // Lyrics with performance notes
}

Performance Notation in Lyrics

"lines": [
  "Main lyric line here",
  "[whispered] Soft delivery",
  "(ad-lib) Yeah! Uh!",
  "[falsetto] High note section",
  "(Marx! Marx!) Background chants",
  "[gasping] Breathless delivery"
]

Style Engineering

The Power of Negative Styles

• Genre bleeding: "negative": ["country", "folk", "jazz"]

• Energy drops: "negative": ["slow", "ballad", "ambient"]

• Vocal issues: "negative": ["monotone", "spoken word", "whispered"]

BPM Sweet Spots

• 70-90: Slow burn, dark, menacing

• 90-110: Hip-hop, trip-hop, groovy

• 120-128: Pop, house, standard dance

• 140: Dubstep, trap, aggressive

• 160-180: Drum & bass, hardcore

Vocal Style Matrix

Avoiding AI Weirdness

❌ COMMON AI FAILS

1. Non-words: "Satisfize", "Chorizo-ing"

2. Awkward scansion: Lines that don't fit the beat

3. Random melodic jumps: Unsingable intervals

4. Energy mismatches: Soft vocals over aggressive beats

✅ PREVENTION STRATEGIES

1. Natural Language Only

// BAD
"Marx-ifying your senses"
"Chorizolicious fever dream"

// GOOD
"Marx is calling out your name"
"Chorizo fever in my brain"

2. Syllable Counting

• 4/4 time = multiples of 4 or 8 syllables work best

• Leave space for breath between phrases

• Test by speaking lyrics in rhythm

3. Concrete Imagery

// VAGUE (AI struggles)
"Feelings of desire"

// SPECIFIC (AI succeeds)
"Behind the freezer door"

4. Musical Directions

"positive_local_styles": [
  "on-beat vocals",      // Prevents off-beat weirdness
  "clear enunciation",   // Prevents mumbling
  "melodic stability"    // Prevents random pitch jumps
]

Advanced Techniques

1. Hook Engineering

• Repetition: Same phrase 2-3 times

• Call-response: Question → Answer pattern

• Sonic branding: Recurring sound/chant (e.g., "Marx! Marx!")

2. Energy Mapping

Intro      ▁▁▂▂   (20% energy)
Verse 1    ▂▂▃▃   (40% energy)
Pre-Chorus ▃▃▄▅   (60% building)
Chorus     ▅▅▆▇   (85% peak)
Verse 2    ▃▃▄▄   (50% energy)
Bridge     ▄▅▆▇   (70-90% climb)
Final      ▇▇██   (100% maximum)
Outro      ▆▄▂▁   (Fade out)

3. Transition Smoothing

"lines": [
  "...",
  "(building to chorus)",  // AI understands this
  "...",
  "[drums intensify]"      // Production cue
]

4. Multilingual Integration

"lines": [
  "English main line",
  "¡Español for emphasis!",
  "(French whisper: oh là là)"
]

5. Genre Fusion Formula

"positive_global_styles": [
  "primary_genre",      // 60% influence
  "secondary_genre",    // 30% influence
  "accent_genre"        // 10% spice
]

Genre Templates

Dark Electronic

{
  "positive_global_styles": [
    "dark techno", "140 BPM", "industrial",
    "heavy bass", "distorted vocals", "menacing"
  ],
  "negative_global_styles": [
    "happy", "major key", "acoustic", "soft"
  ]
}

Hyperpop Chaos

{
  "positive_global_styles": [
    "hyperpop", "150 BPM", "autotuned",
    "glitchy", "chaotic", "maximalist production"
  ],
  "negative_global_styles": [
    "minimal", "organic", "subtle", "relaxed"
  ]
}

Sultry R&B

{
  "positive_global_styles": [
    "R&B", "90 BPM", "sultry vocals",
    "smooth bass", "intimate", "late-night vibes"
  ],
  "negative_global_styles": [
    "aggressive", "fast", "harsh", "childish"
  ]
}

Comedy Rap

{
  "positive_global_styles": [
    "hip-hop", "95 BPM", "comedic delivery",
    "bouncy beat", "clear enunciation", "playful"
  ],
  "negative_global_styles": [
    "serious", "mumble rap", "dark", "aggressive"
  ]
}

Pro Tips from Testing

1. Test Incrementally: Generate 30-second tests before full songs

2. Version Control: Number your JSONs (v1, v2, v3...)

3. A/B Testing: Make two versions with one variable changed

4. Streaming Mode: Use /v1/music/stream for real-time preview

5. Batch Generation: Queue multiple versions overnight

6. Documentation: Comment your JSONs with // (API ignores these)

Troubleshooting

Issue: Melody sounds "off"

Issue: Energy drops mid-song

Issue: Vocals unclear

Issue: Wrong genre bleeding in

Issue: Boring/repetitive

Quality Checklist

Next-Level Composition Psychology

• Confidence: Bold, declarative style descriptions work better

• Specificity: "125 BPM" > "fast tempo"

• Cultural references: It knows genre conventions

• Production terminology: Use real music production terms

• Emotional clarity: One clear mood > mixed emotions

How to Write Songs That Don't Sound AI-Generated

The Uncanny Valley of AI Music

What Makes a Song Sound "AI"?

1. Prosody Violations - Words that don't match the rhythm

2. Semantic Drift - Lyrics that lose coherent meaning

3. Emotional Inconsistency - Mood swings that don't make sense

4. Melodic Randomness - Notes that don't follow musical logic

5. Energy Mismatches - Production that fights the vocals

Anti-Pattern #1: The Word Salad

❌ AI-Generated Garbage

"Mystical sensations flowing through existence
Dancing particles of emotional persistence
Universe calling with synthetic dreams
Reality fracturing at quantum seams"

✅ Human-Sounding Alternative

"Late night in your apartment
City lights through the window
You're dancing in the kitchen
To a song on the radio"

Anti-Pattern #2: The Rhythm Wrecker

❌ Broken Scansion

"I'm desperately wanting to find you" (10 syllables)
"Come to me" (3 syllables)
"The universe is calling out for our love" (11 syllables)
"Yeah" (1 syllable)

✅ Rhythmically Consistent

"I've been searching for you" (6)
"Come and find me too" (5)
"Every star above us" (6)
"Knows our love is true" (5)

Anti-Pattern #3: The Fake Slang

❌ AI Trying to Be Cool

"Vibing on that fleek sauce
Getting all algorithmically boss
My neural network's on fire
Quantum entangled desire"

✅ Actual Slang That Works

"Got me feeling some type of way
Your moves are crazy, I must say
This energy's off the charts
You're playing games with my heart"

Anti-Pattern #4: The Mood Whiplash

❌ Emotional Chaos

Verse: "I'm so depressed and lonely"
Chorus: "PARTY TIME! WOO! YEAH!"
Bridge: "Contemplating existence..."
Outro: "RAGE AGAINST THE MACHINE!"

✅ Emotional Journey

Verse: "Starting to feel the rhythm"
Chorus: "Now we're dancing freely"
Bridge: "This moment's everything"
Outro: "Never want this night to end"

The Comedy Problem

How to Be Funny Without Being Cringe

❌ AI "Humor"

• Random references ("Banana hammock Tuesday!")

• Forced wordplay ("Meat me at the meat meet")

• Over-explaining jokes ("That's funny because...")

✅ Actually Funny

• Unexpected truth ("Your mom likes my playlist")

• Clever innuendo (implied, not stated)

• Callback humor (reference earlier lines)

• Situational absurdity (believable but weird)

The Innuendo Matrix

Subtlety Levels

Double Meaning Masterclass

Song Structure Psychology

The Attention Curve

0-8 sec:    Hook them (MUST grab attention)
8-20 sec:   Set scene (establish world)
20-45 sec:  Build tension (create need)
45-60 sec:  Release (satisfy with chorus)
60-90 sec:  Develop (add complexity)
90-120 sec: Climax (peak energy)
120+ sec:   Resolution (satisfying end)

The Energy Map That Works

Intro:      ████░░░░░░ 40%
Verse 1:    ██████░░░░ 60%
Pre-Chorus: ███████░░░ 70%
Chorus:     █████████░ 90%
Verse 2:    ██████░░░░ 60%
Chorus:     █████████░ 90%
Bridge:     ████████░░ 80%
Final:      ██████████ 100%
Outro:      ████░░░░░░ 40%

Lyrical Frameworks That Never Fail

1. The Story Arc

• Setup: Where/Who/When

• Conflict: The problem/desire

• Rising: Things intensify

• Climax: Peak moment

• Resolution: How it ends

2. The List Song

• Verse 1: List examples

• Chorus: The main point

• Verse 2: More examples

• Bridge: The twist

3. The Conversation

• Verse 1: You said...

• Verse 2: I said...

• Chorus: We both know...

• Bridge: But really...

4. The Progression

• Verse 1: Beginning

• Verse 2: Middle

• Bridge: Transformation

• Final: End state

Vocal Delivery Specifications

What Actually Works

"[whispered]"     - Start of intimate sections
"(ad-lib: yeah)"  - Between main lines
"[falsetto]"      - Emotional peaks only
"(harmony: ooh)"  - Background vocals
"[spoken]"        - Breakdowns, not verses

What Sounds Robotic

"[randomly yelling]"
"(constant ad-libs every line)"
"[switching delivery mid-word]"
"(unclear what this means)"

The Spanish Meat Cookbook 🌶️

Actual Spanish Meats (Use These)

• Chorizo - Spicy sausage

• Jamón ibérico - Premium ham

• Morcilla - Blood sausage

• Lomo - Pork loin

• Cecina - Cured beef

• Sobrasada - Spreadable sausage

• Butifarra - Catalan sausage

Temperature/Preparation Terms

• "Sizzling" > "Hot meating"

• "Grilled to perfection" > "Meat-ified"

• "Slow-cooked" > "Meat processing"

• "Marinated" > "Meat-soaked"

Marx Foodservice Specific Guidelines

What Works

• Marx as a chant/hook

• Specific location references (break room, freezer)

• Actual food service terminology

• College setting details

• Workplace hierarchy humor

What Doesn't

• Over-explaining the company

• Making up food service terms

• Forcing "Marx" into weird compounds

• Generic workplace references

Testing Your Lyrics

The Speak-Aloud Test

1. Read your lyrics out loud

2. Do they sound like something a human would say?

3. Can you say them in rhythm?

4. Do they make sense without music?

The Cringe Test

1. Would you be embarrassed if someone found these lyrics?

2. Would a stranger understand the humor?

3. Is the innuendo clever or just vulgar?

4. Would this work at a party?

The Energy Test

1. Map energy levels 1-10 for each section

2. Does the progression make sense?

3. Are transitions smooth?

4. Is the climax actually climactic?

Common Fixes

Problem: "It sounds like a robot wrote this"

Problem: "The rhythm is off"

Problem: "It's not funny"

Problem: "Too vulgar/obvious"

Problem: "Energy is flat"

The Ultimate Checklist