Skip to content

Latest commit

 

History

History
723 lines (513 loc) · 28.8 KB

File metadata and controls

723 lines (513 loc) · 28.8 KB

Suno V5 / V5.5 Best Practices

Comprehensive guide for getting the best results with Suno V5 and V5.5.

Related skills: /bitwize-music:suno-engineer (interactive prompting), /bitwize-music:pronunciation-specialist (phonetic review) Related docs: pronunciation-guide.md, structure-tags.md, voice-tags.md, tips-and-tricks.md


V5.5 Update (March 26, 2026)

V5.5 is an evolution of V5, not a break from it. Prompt syntax, metatags, structure tags, creative sliders, the 1,000-character style box, and the 5,000-character lyrics box are all unchanged — V5 prompts run identically on V5.5. No patterns are deprecated.

What changed is engine responsiveness and personalization:

Change Impact on prompting
Nuanced phrasing, stronger dynamic range Worth trying finer descriptors (e.g., "slightly detuned vintage keys") — improved nuance is engine-reported, not independently verified
Better instrument separation Busy arrangements stay readable — less mud on dense prompts
More expressive vocals Emotion tags (breathy, yearning, resigned) track closer to intent
Voices (Pro/Premier) Voice cloning replaces vocal persona descriptors — see below
Custom Models (Pro/Premier) Fine-tuned model carries style — style prompts can be shorter
My Taste (all tiers) Passive preference learning; affects style autogenerate, not explicit prompts

Practical guidance: keep doing what V5 prompting teaches. If anything, trust the engine more — a touch less over-specification, a touch more reliance on one or two evocative descriptors.

See the Voices & Custom Models section below for the V5.5-only features.

Quick Start Formula

[genre], [subgenre], [instruments], [mood], [tempo], [vocal description]

Example:

nerdcore hip-hop, glitchy IDM beats, lo-fi digital artifacts,
nostalgic, melancholic, 85 BPM, male vocals, gravelly voice, introspective

V5 Key Improvements

Feature Description
Intelligent Composition Coherent structure from 30-second hooks to 8-minute epics
Studio-Grade Audio 44.1 kHz output with fuller, more balanced mixes
Vocal Engine Human-like vocals with breath, emotion, vibrato control
10x Faster Seconds instead of minutes for generation
Stem Separation (3 modes) Auto Split (12 stems), Split from Mix (target + the rest), Advanced Split (~100 instruments) — generative regeneration
Extended Length Up to 8 minutes per generation
Persistent Memory Vocal characters and instruments remain stable across project generations
Granular Controls Tempo, key, dynamics, arrangement with optional automation

Critical Rule: Don't Reuse Old Prompts

Suno CTO's #1 recommendation: Don't rerun old V4/V4.5 prompts on V5.

V5 listens differently and needs less instruction. Write new prompts and experiment.


Prompt Construction

Keep It Simple — Avoid Prompt Fatigue

V5 is literal, and it dilutes attention when descriptors repeat the same idea. Every descriptor should earn its place — genre, instrument, vocal identity, production texture, mood, tempo. A focused style box of ~10 descriptors works well; what hurts is a synonym-pile (five mood words that all mean "soft") that gives V5 nothing new to act on.

On the "4–7 descriptors" rule of thumb: 4–7 is a useful starting point, not a Suno-official rule. Rich ~10-descriptor style boxes are common and effective when every term does distinct work. Trim synonyms, not detail. (The hard "cut everything past 7" version of this rule traces to a single third-party guide and isn't borne out in practice — real style boxes routinely run richer.)

❌ Bad (synonym-pile — overlapping mood words that add nothing new):
"Ethereal indie folk with vintage analog warmth and melancholic
undertones, finger-picked acoustic, tape hiss, lo-fi, intimate,
breathy, whispery, nostalgic, contemplative, minimalist production"

❌ Bad (too vague):
"Nice upbeat music"

✅ Good (every descriptor pulls its weight):
"Sad indie folk, acoustic, gentle, breathy female vocal, intimate"

Rule of thumb: if descriptors start restating the same idea (intimate / breathy / whispery / soft), collapse them into one. Don't pad — but don't strip out genuinely distinct instrument, vocal, or production detail just to hit a number.

The Four-Part Anatomy

1. Genre + Era + Influences
   "90s alt-rock with Britpop undertones"

2. Tempo/BPM + Key (optional)
   "120 BPM, A minor"

3. Instrumentation & Arrangement
   "Live drums with room ambience; palm-muted guitars; warm bass"

4. Production & Mix Notes
   "Analog glue compression; tape saturation; lead vocal upfront"

Alternative: Top-Loaded Palette Formula

A simpler approach that front-loads the most impactful elements:

[Mood] + [Energy] + [2 Instruments] + [Vocal Identity]

Example:

Melancholic, slow-burn, piano and strings, female alto with subtle vibrato

Genre-Specific Tips

Hip-Hop / Rap

  • Specify subgenre: boom bap, trap, lo-fi, nerdcore
  • Include beat style: 808s, sampled drums, crispy snares
  • Describe flow if important

Punk

  • Specify subgenre: pop-punk, hardcore, skate punk
  • Note tempo (punk is usually fast)
  • Describe vocal style: snotty, shouted, melodic

Electronic

  • Name specific subgenres: house, techno, IDM, synthwave
  • Describe synth types: analog, digital, chiptune
  • Include BPM (critical for dance music)

Folk/Acoustic

  • Specify instruments: fingerpicking, banjo, mandolin
  • Note tempo and mood
  • Describe vocal intimacy level

K-Pop

K-pop presents unique challenges for Suno due to its maximalist production, multi-vocal architecture, language mixing, and mid-song genre shifts.

Core style prompt approach:

  • Always include K-pop explicitly plus specific production terms (maximalist, glossy, dynamic shifts)
  • Place vocal description first: mixed group vocals, layered harmonies, K-pop idol group
  • Specify concept type: girl crush, cute concept, dark concept, retro disco
  • Include BPM (dance tracks: 120-140, ballads: 60-80, hip-hop: 80-100)

Getting the group vocal sound:

  • mixed group vocals is the single most important tag for simulating a multi-member group
  • Add layered harmonies, gang vocals, group chant for chorus density
  • Use parenthetical backing vocals in lyrics: I'm on fire (on fire!) to trigger echo/response
  • Use [All] or [Group] section tags before chant sections
  • Separate rap verses with [Rap Verse] tags to signal a different vocal character

Korean-English code-switching:

  • Suno V5 handles Korean (Hangul) better than earlier versions, but romanized Korean with hyphens remains more reliable for pronunciation control
  • Format: Sa-rang-hae not Saranghae
  • Add [Clear Vocals] or [High Fidelity Vocals] when mixing languages
  • Keep English hooks and Korean verses as separate sections when possible

The "switch-up" (mid-song genre change):

  • Use parenthetical genre cues at each section, not just in the global style prompt:
    [Verse 1]
    (Soft R&B groove, gentle piano)
    ...lyrics...
    
    [Chorus]
    (Explosive EDM drop, heavy bass, full energy)
    ...lyrics...
    
    [Rap Verse]
    (Aggressive trap flow, 808 bass)
    ...lyrics...
    
    [Bridge]
    (Stripped-back ballad, solo piano)
    ...lyrics...
    
  • Include genre-fluid, dynamic shifts, maximalist K-pop production in style prompt

K-pop section structure (more sections than Western pop):

  • Intro → Verse 1 → Pre-Chorus → Chorus → Post-Chorus → Verse 2 → Pre-Chorus → Chorus → Rap Verse → Bridge → Dance Break → Final Chorus (key change up) → Outro
  • Use [Dance Break] with (Instrumental, heavy beat) and minimal/no lyrics
  • Final chorus often modulates up a half-step — note (key change up, maximum energy) in parenthetical

Common K-pop Suno issues:

Problem Solution
Sounds like generic pop, not K-pop Add K-pop + maximalist + glossy + specific concept keywords
Solo voice instead of group Add mixed group vocals, layered harmonies, K-pop idol group
No genre switch-up mid-song Use parenthetical genre cues per section, not just global prompt
Korean pronunciation garbled Use romanized Korean with hyphens; add [Clear Vocals]
Rap verse sounds same as singing Use [Rap Verse] tag with (aggressive rap flow) parenthetical
Dance break has singing Keep lyrics minimal/empty in [Dance Break]; add (Instrumental)

Example K-pop style prompts by concept:

Girl crush:

K-pop girl group, fierce EDM trap, sassy vocals, heavy bass drop, chant chorus, 135 BPM, confident attitude, glossy production

Bright/cute:

K-pop, bubblegum pop, bright synths, chirpy vocals, catchy hook, youthful energy, 125 BPM, layered harmonies

Dark/experimental:

K-pop, industrial synths, aggressive rap, EDM bass drops, distorted bass, maximalist chaos, 140 BPM, mixed group vocals

K-ballad:

Korean ballad, emotional piano, string orchestra, soaring vocals, key change final chorus, 70 BPM, cinematic, lush arrangement

Vocal Control

Top-Anchor Approach

Start your prompt with vocal description before lyrics:

Female pop vocalist, breathy, intimate, 90s R&B groove

[Verse 1]
Lying in the dark tonight...

Vocal Persona Examples

Male tenor, warm, slightly raspy, indie rock delivery
Female alto, sultry, breathy, R&B phrasing with subtle runs
Male baritone, gravelly, introspective, folk storyteller

Section-by-Section Dynamics

Section Dynamics Phrasing Vibrato
Verse Low Tight Minimal
Pre-Chorus Rising Shorter Growing
Chorus High/Open Sustained Full
Bridge Variable New texture Altered

Lyric Formatting

Keep Lyrics Concise

Shorter lyrics generate better results. Dense or long lyrics cause Suno to rush, compress sections, or skip content entirely. Target 200–350 words for most genres (up to 500 for hip-hop/rap). Two verses plus chorus plus bridge is the sweet spot — avoid 4–5 verse songs.

Use Explicit Section Tags

[Verse 1]
Walking through the rain tonight
Memories fading out of sight

[Pre-Chorus]
But I still remember when

[Chorus]
We were young and free

Sound Effects

Trigger vocal sound effects by placing them in brackets:

[Verse 1]
Walking through the night [footsteps]
I hear a voice calling [echo]
Then suddenly [laughter] breaks the silence

Common Effects:

Human:

  • [laughter] - Natural laughing
  • [screaming] - Vocal scream
  • [whisper] - Whispered delivery
  • [sigh] - Breathing out, reflective
  • [gasp] - Sharp intake of breath
  • [cough] - Clearing throat, realistic aside

Crowd:

  • [crowd] - Crowd noise
  • [applause] - Clapping/applause
  • [cheering] - Crowd excitement

Mechanical/transition:

  • [echo] - Echo/reverb effect
  • [phone ringing] - Telephone sound, narrative device
  • [static] - Radio noise, transition
  • [record scratch] - DJ/hip-hop transition marker

Note: Effects work best when placed mid-line, not as standalone lines

Atmospheric Effects

For environmental sounds (rain, wind, fire), mention in both the Lyrics Box and Style Prompt:

Lyrics Box:

[Verse]
Rain falling on the window
Thunder in the distance

Style Prompt:

lofi effects rain, ambient thunder

Why Both?: Repetition strengthens AI recognition of desired atmosphere

Common Atmospheres:

  • rain + "lofi effects rain" (style prompt)
  • wind + "ambient wind textures" (style prompt)
  • fire + "crackling fire ambience" (style prompt)
  • ocean + "ocean waves background" (style prompt)

Syllable Control

  • Verse lines: 7–9 syllables per line for best vocal lock-in
  • Chorus lines: 10–12 syllables per line
  • Overall range: 6–12 syllables per line
  • Use hyphens for sustained notes: lo-ove, sooo-long
  • Writing Loooove or Ohhhh creates sustained notes and vocal emphasis
  • ALL CAPS can create a shouting effect: NEVER AGAIN
  • Punctuation signals phrasing: commas = pauses, ellipses = trailing

Prevent Lyric Alterations

Add at the top of your prompt:

Do not change any words. Sing exactly as written.

Negative Prompting

Exclusions shift the odds against an element — they're probabilistic, not a hard filter, and won't override a prompt that strongly implies the thing you're excluding.

Two ways to exclude:

  • Dedicated Exclude Styles field (Custom Mode → Advanced Options, Pro/Premier) — the reliable path. Put exclusions here, not buried in the main style prompt.
  • Inline no [element] appended to the style box — the fallback when you don't have the field (free tier). Weaker; the engine may still slip the element in.

Keep it to 2–4 items — over-specifying dilutes the effect.

What You Can Exclude

  • Instruments: "no drums", "no electric guitar"
  • Vocal effects: "no autotune", "no heavy reverb"
  • Stylistic elements: "no EDM drops", "no screaming"
  • Unwanted group vocals (a common Suno over-add): "no choir", "no crowd vocals", "no backing vocals", "no gang vocals", "no call-and-response", "no vocal harmonies", "no layered vocals"

Best Practices

✅ Good:
"Acoustic folk, warm, intimate, no drums, no electric instruments"

❌ Bad (over-specified):
"No drums, no bass, no synths, no reverb, no distortion, no..."

Group vocals are probabilistic to suppress: excluding "choir / crowd / backing vocals" improves your odds, but a big anthemic prompt can still pull them back in. Pair the exclusion with a leaner, more intimate style prompt for the strongest effect.


Bar Count Targeting

V5 supports targeting specific bar counts per section using numbers after section tags:

[INTRO 4] [VERSE 1 8] [PRE 4] [CHORUS 8] [VERSE 2 8] [PRE 4] [CHORUS 8] [BRIDGE 8] [CHORUS 8] [OUTRO 4]

The numbers represent target bar counts for each section. This gives you finer control over song structure and pacing beyond just section tags.

Notes:

  • Results are approximate — Suno treats these as targets, not guarantees
  • Combine with explicit section tags in the lyrics box for best results
  • Works well for controlling intro/outro length

Creative Sliders

V5 includes sliders in the generation interface that affect output:

Slider Effect Guidance
Weirdness Higher = more experimental and unexpected choices Raise to explore; lower for predictable, hooky results
Style Influence Higher = tighter adherence to style prompt Raise for genre purity; lower for looser fusions
Audio Influence Controls how much uploaded audio shapes the output Appears only when audio is uploaded

Tips:

  • Start with default values and adjust after hearing the first generation
  • High Weirdness + specific genre tag = interesting results within a genre
  • Low Style Influence is useful when you want the AI to surprise you

Deep dive: creative-sliders.md — per-slider behavior, genre starting ranges, interaction effects, and when to move a slider vs. rewrite the prompt.


Voices & Custom Models

V5.5 only. Pro and Premier subscribers.

Voices (voice cloning)

Upload a clean acapella, a full track with background music, or sing directly into a mic (15 seconds to 4 minutes of material — the cleaner, the less needed). Suno then has you read a random spoken phrase aloud and matches it to the uploaded audio as a consent/ownership check. Cloned voices are private to the account; sharing is announced but not yet live.

  • Cost: 4 credits per creation (beta pricing).
  • Consent box is mandatory — activating Voices grants Suno permission to use your voice data to train their models broadly, not just your private instance. This is not optional for activation.
  • Age-gated: 18+.

Prompting with a Voice:

  • Drop gender/register descriptors from the style box — the Voice carries them. Free that budget for genre, instrumentation, and mood.
  • Voice + Persona is redundant; pick one.
  • Keep the style prompt to 1–2 genres plus instrumentation, same as the Personas rule.

Custom Models (fine-tuning)

Upload at least 6 original tracks from your catalog. Suno fine-tunes a private V5.5 on your harmonic preferences, arrangement habits, instrumentation choices, and production aesthetic. Build time: 2–5 minutes. Up to 3 models per account, maintained concurrently.

Prompting with a Custom Model:

  • Drop generic production language ("glossy", "modern pop production", "polished mix") — the model already encodes your aesthetic.
  • Keep genre and section-level direction. Specific one-off choices (tempo, featured instrument, mood shift) still matter.
  • Best for series/album consistency. Generic v5.5 is often better for deliberately off-brand tracks.

My Taste (all tiers, including free)

Runs passively in the background, learning genres, moods, and styles from your activity. It shapes the style autogenerate feature — it does not override explicit prompts. No action required, but worth knowing it exists when the autogenerate button starts reading your mind.


Personas

Available to: Pro and Premier subscribers

Personas let you save the "essence" of a generated song — vocals, style, vibe — and reuse it across different songs. This is the most reliable way to maintain vocal consistency across an album.

Creating a Persona

  1. Generate a song with vocals you like
  2. Save the song's vocal identity as a Persona
  3. Apply the Persona to future generations

Best Practices

  • Keep prompts simple when using Personas (1–2 genres). The Persona carries the vocal identity, so you don't need to re-describe the voice.
  • Personas can be dominant — if your style prompt fights the Persona, the Persona usually wins. Work with it, not against it.
  • Voice Personas lock in a specific AI singer independent of musical style. You can move a Persona across genres (e.g., same singer doing folk and electronic).

Limitations

  • 200 free songs with Personas per billing cycle; then 10 credits per song
  • December 2025 update made Personas more dominant in the mix — adjust style prompts if the Persona is overpowering other elements

Song Editor

V5 includes a section-level Song Editor that lets you modify individual parts of a generated song without regenerating the whole track.

Capabilities

Action Description
Remake Regenerate a section with the same prompt
Rewrite Change lyrics/melody for a section while preserving role and intent
Extend Append bars at the tail of a section
Reorder Move sections around in the arrangement
Delete Remove weak sections; transitions are engine-handled

Workflow

  1. Generate a full song
  2. Identify sections that need improvement
  3. Use Remake/Rewrite on individual sections
  4. Extend 1–2 bars into/out of a chorus for smooth transitions
  5. Delete weak regions — the engine handles transition smoothing

Note: Keep extensions to 2–3 times max per song. Extending too many times causes uneven lyrics, weaker vocals, and quality drops.


Token Biases Warning

Suno's model has known token biases — it gravitates toward certain words when generating or interpreting lyrics. These are model preferences, not creative choices:

Common bias words: Neon, Echo, Ghost, Silver, Shadow, Whisper, Crystal, Velvet

If you find these words appearing in your generations when you didn't write them, it's the model defaulting to its favorites. Use the "Do not change any words. Sing exactly as written." instruction to prevent unwanted substitutions.


Troubleshooting

Problem Solution
Vocal too buried Add: "lead vocal 1–2 dB louder than band"
Mix feels flat Add: "bus compression 2–3 dB, slow attack/fast release"
Arrangement too busy Specify rests: "verse 2: bass rests for 4 bars"
Genre drift Reassert influences mid-prompt
Chorus not lifting Add: "double-time hats; octave guitars"

Suno Output Loudness (Pre-Mastering)

These are typical loudness levels Suno generates — not final mastering targets. For streaming platform delivery, master all tracks to -14 LUFS / -1.0 dBTP regardless of genre. See /reference/mastering/mastering-workflow.md for mastering procedures.

Genre Typical Suno Output
Pop/EDM -9 to -7 LUFS
Lo-Fi -12 to -11 LUFS
Podcast/Spoken -16 to -14 LUFS

Iteration Tips

  1. Start broad, then refine
  2. Log every attempt - note what worked/didn't
  3. Adjust one element at a time when refining
  4. Try different models - V4.5 vs V5 produce different results
  5. Use extends to build on good sections

Stem Extraction

Suno's stem separation was overhauled (June 2026) into three selectable modes, and it now generatively regenerates each stem rather than frequency-carving the mix — extracted parts sound cleaner and more natural than the old model.

Split Modes

Mode What it does Output
Auto Split The classic model — splits a song into 12 stem categories at once 12 stems
Split from Mix Pulls one chosen instrument or voice out of the mix 2 stems: the target, and everything-else-without-it
Advanced Split Extracts one specific instrument chosen from a list of ~100 (drum kit to didgeridoo) 1 targeted stem (Premier; per-extraction credit cost)

Auto Split — the 12 stems:

Vocals, Backing Vocals, Drums, Bass, Guitar,
Keyboard, Strings, Brass, Woodwinds,
Percussion, Synth, FX/Other

Extraction Workflow

  1. Click More Actions (...) on any clip
  2. Hover over Get Stems
  3. Choose your split mode — Auto Split (all 12), Split from Mix (one target + the rest), or Advanced Split (one instrument from ~100)
  4. Import into DAW

Cleaner Single-Stem Pulls

For an isolated vocal or instrument, Split from Mix targeting that part is usually cleaner in one pass than the old carve-and-repeat trick (a benefit of the generative model). If a stem still bleeds, run extraction again on the extracted stem.

Tier note: Advanced Split is Premier-tier with a per-extraction credit cost (exact costs vary — check Suno's current pricing). Auto Split / Split from Mix availability follows your plan.


Suno Studio (Premier Plan)

Released: September 25, 2025 Availability: Premier plan required

Suno Studio is a generative audio workstation that combines AI music generation with professional editing tools.

Key Features

Feature Description
Multitrack Editor Timeline-based editing with drag-and-drop
Stem Controls Generate, separate, and manipulate individual tracks
MIDI Export Export compositions as MIDI for DAW integration
Audio Upload Import existing audio and manipulate with AI
Sample to Song Upload short snippets and expand to full compositions
Pitch Transpose Adjust pitch by semitones without regenerating

Sample to Song Workflow

  1. Click Upload in Suno Studio
  2. Select a short audio file (guitar riff, vocal melody, etc.)
  3. Describe the desired full composition in the prompt
  4. Suno expands the snippet into a complete track
  5. Edit on timeline, adjust stems, export MIDI

Use Cases:

  • Record guitar lines and build full arrangements around them
  • Capture vocal ideas and develop into complete songs
  • Import samples and integrate into AI-generated tracks

Pitch Transposition

Access: Remix → Suno Studio (Premier plan)

  1. Click generated song on timeline
  2. Locate transpose slider under clip settings
  3. Adjust pitch by semitones (±12 range)
  4. Preserves melody, phrasing, and rhythm

Benefit: Fix key mismatches without wasting credits on regeneration


Known V5 Limitations

  • Heavy electric guitars can sound "dirty" or blend together
  • Acoustic nuance not always captured perfectly
  • Niche subgenres (metalcore, extreme styles) may miss hallmarks
  • Extreme cross-style fusions → muddy results
  • Quality may degrade past 6-7 minutes
  • V4.5 may produce better results for heavy genres (metal, hardcore) — consider testing both if V5 output sounds thin

Ownership & Licensing (Post-November 2025)

Following the Warner Music Group partnership (November 2025):

  • Subscribers get "commercial use rights" but are "generally not considered the owner" of generated content
  • Suno will not take a revenue share from monetization
  • Artists/songwriters retain full control over name/image/likeness/voice use
  • New models trained on licensed WMG catalog are planned for 2026
  • Current models will be deprecated when licensed models launch — download your catalog

Action item: Download all important generations now. Current-model content may become inaccessible when licensed models roll out.


Quick Reference Card

PROMPT TEMPLATE:
[Genre], [BPM], [mood/vibe]
[Vocal]: [gender], [texture], [style]
[Instruments]: [2-4 key instruments]
[Mix]: [1-2 production hints]

STRUCTURE TAGS:
[Intro] [Verse] [Pre-Chorus] [Chorus]
[Bridge] [Breakdown] [Outro] [End]

VOCAL TAGS:
breathy, raspy, powerful, intimate, ethereal,
gravelly, smooth, aggressive, tender, soulful

MIX TAGS:
punchy, wide stereo, vintage, modern, lo-fi,
crisp, warm, bright, deep, spacious

Related Skills

  • /bitwize-music:suno-engineer - Technical Suno V5 prompting expert

    • Uses this guide as reference
    • Constructs style prompts and genre tags
    • Optimizes prompts for best generation results
  • /bitwize-music:lyric-writer - Lyric writing with Suno formatting

    • Automatically formats lyrics with section tags
    • Prepares Suno-ready lyrics boxes
    • Applies pronunciation fixes for Suno
  • /bitwize-music:lyric-reviewer - Pre-generation QC gate

    • Verifies lyrics follow Suno best practices
    • Checks section tags and structure
    • Ensures lyrics are ready for generation

See Also

  • /reference/suno/creative-sliders.md - Weirdness / Style Influence / Audio Influence deep dive: genre starting ranges, interaction effects, slider-vs-prompt
  • /reference/suno/pronunciation-guide.md - Phonetic spelling, homographs, pronunciation fixes
  • /reference/suno/structure-tags.md - Complete list of section tags ([Verse], [Chorus], etc.)
  • /reference/suno/genre-list.md - 500+ genre tags for style prompts
  • /reference/suno/voice-tags.md - Vocal style descriptors and tags
  • /reference/suno/tips-and-tricks.md - Troubleshooting, extending tracks, operational tips
  • /skills/suno-engineer/SKILL.md - Complete Suno engineer skill documentation

Sources