| Предыдущая глава |
↓ Содержание ↓
↑ Свернуть ↑
| Следующая глава |
For vocals specifically, tags like [Male], [Female], [Choir], [Duet] (and their v4.5 variants [Male Vocal], etc.) are crucial. They tell Suno what kind of voice to use. In a song context, you might start your lyrics with [female:] or simply tag the chorus with [choir] to add choral backing. Suno v4.0 did allow this, but v4.5 expanded the realism and range. For example, v4.5 can produce richer choir harmonies when given [Choir] — it understands parameters like "angelic" or "powerful" on that tag — whereas v4.0’s choir might have been more muted. The documentation lists detailed parameters for [choir] (layered, gospel, dissonant, etc.) which presumably apply to both versions, but v4.5 has the edge in executing those styles convincingly.
Another vocal-centric tag is [harmonies] / [background-vocals], which instructs adding harmony vocals behind the lead. In songs, this is widely used for choruses or to thicken important lines. The doc’s [background-vocals] tag (with parameters like harmonic, layered) is clearly for vocal tracks. Suno v4.0 could add background "oohs" or harmonies, but sometimes it struggled to keep them musically aligned. V4.5 does a better job — for instance, if you specify [background-vocals: layered harmonies in the chorus], v4.5 is more likely to produce a satisfying result (stacked harmonies on the chorus lines) than v4.0 was. This aligns with v4.5’s improved "richer vocals" claim.
Spoken segments and effects: In vocal songs, tags like [Spoken Word], [Rap] (if used) or things like [laugh], [sigh] (sound effect tags for vocals) come into play. The documentation included [laughter] as a tag for adding laughter SFX, which is obviously only relevant if vocals are present. Both v4.0 and v4.5 can inject these vocal effects if prompted (though these are more for fun — a well-placed evil laugh in a metal song, etc.). There is also a [whisper] tag in the doc, denoting a whispered vocal style. This is clearly vocal-oriented; you’d use it in a lyric line or section to get a whispery delivery. Notably, v4.5 handles whispered and spoken vocals more deftly — earlier versions sometimes sang the words anyway or made them too loud. V4.5, if given [whisper: ...] at the intro or verse, will often produce actual whispering voices, which adds a creepy or intimate effect as intended. This is another area where the tag’s behavior improved with the version.
Finally, persona/voice tags: In vocal songs, controlling the character of the voice is key. We discussed how [personae:] tag was a user-documented attempt to lock a voice persona (gritty male, etc.), but it’s not officially supported due to the Persona feature being separate. Instead, users achieve consistent voice by tagging the first verse or chorus with gender/range and tone (e.g. [Male: gritty baritone]). Both v4.0 and v4.5 allow that kind of description. V4.5 introduced the official Personas UI where you can simply select a voice profile (which was separate from the text prompt). If a Persona is selected, it essentially overrides what a meta-tag might try to do. So one could say a version-dependent behavior is: in v4.5, if you have a Persona chosen, tags like [Male Vocal] or [Female Vocal] might be redundant or ignored (because the model is already locked to a specific voice). In v4.0 (which had no Personas feature), those tags were the only way to specify voice gender. So usage shifts: v4.0 relied on meta-tags for voice selection, while v4.5’s UI offers Personas that achieve the same without tags — though you can still use tags if you want to mix multiple voices in one song (like a duet).
In summary, instrumental tracks tend to use tags focusing on instruments, solos, and overall structure, whereas vocal tracks use an additional layer of tags for voice type, lyrics structuring, and vocal effects/harmonies. Both v4.0 and v4.5 share the tag set, but v4.5 executes these with more fidelity (e.g. better separation of an instrumental break, more realistic backing vocals, clearer distinction between a verse and chorus as per tags). When crafting prompts, one should include tags appropriate to the content: for an instrumental, you’d definitely include [instrumental] and instrument tags; for a song, you’d use section and vocal tags (and not include [instrumental], since that would suppress vocals entirely). The good news is Suno v4.5 is versatile enough that you can even combine them — for example, some advanced prompts have a song with an instrumental intro: they literally write something like:
[Instrumental intro]
(guitar chords play, no vocals)
[Verse 1]
Lyrics start here...
And it works (the model starts with an instrumental intro then brings in vocals). This kind of nuanced control is exactly what meta-tags enable, and v4.5 handles it more "intelligently" than earlier versions.
Newly Confirmed or User-Tested Meta-Tags (v4.0 / v4.5)
[vocalist: ]
Purpose: Defines voice identity or texture. Works alongside [vocals].
Example: [vocalist: breathy female mezzo-soprano with jazz phrasing]
[male vocal], [female vocal]
Purpose: Forces male or female vocal presence.
Example: [female vocal] or [male vocal]
[duet]
Purpose: Indicates a two-voice arrangement (often alternating or harmonizing).
Example: [duet: female lead with whispered male counterline]
[spoken word]
Purpose: Triggers narration or monologue mode instead of singing.
Example: [spoken word: distorted narration under ambient pad]
[harmonies: ]
Purpose: Adds stacked or layered vocals; works for both lead and background.
Example: [harmonies: sustained high-third vocal stack under main melody]
[vulnerable vocals: ]
Purpose: Defines tremulous or emotionally fragile vocal tone.
Correct Usage: [vulnerable vocals: trembling voice, cracked falsetto in chorus]
Avoid: Placing descriptive text outside tag.
Instrument Solo Tags
Usable as structural tags, often treated like [section] but with strong instrument focus:
[guitar solo: blues-style run with wah FX]
[sax solo: late-night echo sax with heavy reverb]
[violin solo: baroque trills and glissando]
[synth solo: retro wave arpeggios rise into high delay]
[flute solo: airy modal runs]
[era: ]
Purpose: Suggests period aesthetics, influences instrument and mix style.
Example: [era: early 2000s UK garage]
Expanded [style:] Values (freeform, observed functional)
retro-horror
psychedelic-swing
neo-folk-electronica
horror-synth-cabaret
trap-fugue
sacred-jazz-chant
industrial-surf
musique-concrète-pop
Confirmed Free-Form [style] or [genre] Examples
horror-synth, ghost-folk, spacewestern-phonk, ambient-baroque, baroque-opera
glitch-jazz, vaporwave-trap, noir-hip-hop, musique-concrète, hauntology
enka-minimal-techno, dub-clockpunk, phonk-noir, echo-chamber-pop
Update for v4.5, v4.5+ and upcoming v4.6 (added: August 25, 2025)
New/confirmed tags
New tags confirmed:
aria-rise
build
chant-loop
inversion
lament
polyphony
scat break
subject
technique
New for the current version (v4.5 and later):
hook
rapped verse
distorted vocals
quiet arrangement
Tags obsoleted or changing the behavior
Meta-tags related to audio engineering (e.g. [mix], [master], [filter], [panning], [volume]) are considered ineffective in current Suno versions.
Several early meta-tags have been deprecated in favor of newer equivalents. For example, [sing-style] (from Suno v3) was replaced by [vocal-style] and is ignored in v4. [song-type] (intended to denote song vs. rap vs. instrumental) was an experiment that became inert by v4.0. Likewise, an unlabeled [theme] tag (without a letter or description) no longer works — users must use specific section labels like [Theme A], [Theme B] etc. or it will be ignored. In summary, many v3-era tags were phased out and v4.5 will simply ignore any of these obsolete tags if used.
The catch-all [section: ...] tag has been rendered redundant. Instead of using [section] as a placeholder, the model expects explicit structural tags (intro, verse, chorus, etc.). In fact, any unrecognized tag word is now parsed as a section label by default. Users have found that using the actual section names yields better results, whereas a raw [section: X] might be ignored or misinterpreted. Suno’s documentation explicitly lists section as removed/redundant.
There is no supported [loop] meta-tag to force looping playback. Earlier guides suggested tags like [loop] or [loop chorus], but these do not function in v4.5. Indeed, “loop” is on the removed-tags list. To create a loop or repeated section, users must manually copy structures (or use the Extend feature); a single tag will not make the song endlessly loop.
The autotune effect tag is effectively deprecated in v4.5. While some community guides still mention using [Autotune] for a pitch-corrected vocal stylereddit.com, the official word is that autotune was removed as an unstable tag. Users confirm that simply adding [Autotune] in lyrics has little to no effect now — you may get an “auto-tuned” feel only by describing it in the style text or using a persona that implies it.
The tag [end] (intended to mark the song’s conclusion) exists, but community feedback indicates it’s not very reliable. One user noted that “[end] frequently does not work… even [5 second fade out][end] doesn’t seem to stop the music”, though it does tend to prevent any new vocals after that pointreddit.com. In practice, Suno might ignore an end tag and continue the instrumental to the full length. As a workaround, some creators include an outro section description (and sometimes silence) to encourage a proper ending, since the [end] tag alone is hit-or-miss.
Tags with new behavior
Genre Mashups in [genre:] — Suno v4.5 dramatically improved its handling of combined genres. The genre tag now accepts hybrid values and actually produces blended styles, which older versions often failed to do. For example, a prompt with [genre: midwest emo + neosoul] will yield a coherent mix of those genres in v4.5suno.comsunnoai.com, whereas v4.0 might have defaulted to one genre or produced a muddled result. The model was expanded to recognize 1,200+ genres/styles and interpret “X + Y” combinations smoothlysunnoai.com. This means users can get creative with genre tags (even inventing combos like “jazz-house,” “folk EDM,” “punk meets classical”) and expect v4.5 to honor both elements more faithfully than before.
Richer Tempo Descriptors — The [tempo:...] tag became more nuanced in v4.5. While it still doesn’t support exact BPM numbers, it now understands descriptive tempo phrases much better. In v4.0 one might only use simple terms ([tempo: slow] or fast), but v4.5 can parse complex inputs like “mid-tempo 90s hip-hop swing” or “steady 4/4, 120bpm feel” embedded in a tempo tag. The engine won’t lock to an exact BPM, but it will interpret relative tempo and rhythmic feel from natural language. Essentially, v4.5’s broader language comprehension lets you be more specific in tempo/mood tags (e.g. “slow-burning waltz tempo”) and get a correspondingly specific output, which was less true in earlier versions.
More Expressive Vocal Tags — Tags controlling vocal style/timbre respond more deeply in v4.5. The model now differentiates subtle vocal instructions: for instance, [vocal-style: whispered, airy] or [vocals: nasal, twangy tone] will noticeably affect the performance. In v4.0, many such modifiers were ignored unless they were very common adjectives. Now, however, v4.5 was tuned for emotional and tonal nuance — users report that specifying a singer’s tone (raspy, operatic, whispered, etc.) yields a clear change in the output. Even without a dedicated tag, putting a descriptor in brackets (e.g. [whisper voice]) can work, but the recommended approach is to use the proper tag syntax (like vocal-style or include it in a [vocals: ...] list). Result: a richer palette of vocal textures — from smooth crooning to rough growls — can be invoked via tags in v4.5, whereas previously the model often defaulted to a generic voice.
Strict Instrumental Tag Adherence — The [instrumental] tag (to generate music with no vocals) is honored more reliably post-v4.5. In v4.0 the “instrumental” tag sometimes wasn’t 100% respected (users occasionally heard stray humming or ooohs). In v4.5, by contrast, if you start your lyrics with [instrumental], the model will produce a purely instrumental track almost every time. The upgrade in prompt fidelity means the system correctly silences vocals when asked. Users have also learned they can combine this tag with a brief style hint (e.g. [instrumental: lo-fi hip hop beat]) to guide the instrumental’s genre. This stricter obedience to the instrumental directive is a quality-of-life improvement noted by many after the v4.5 update.
Longer Structures Now Possible — With v4.5’s extended max song length (~8 minutes), the classic structure tags like [Verse], [Chorus], [Bridge], etc., can be used in more repeated sections without being dropped. Previously, a very long prompt with many sections might see later tags get ignored as the model ran out of time/attention (v4.0 often wouldn’t reliably include a 3rd or 4th verse). Now, v4.5 can execute a full song structure end-to-end with multiple verses, choruses, a bridge, even multiple themes, and an outro. Community feedback around the v4.5+ release confirmed that the model maintains coherence over longer sequences, so tags defining, say, Verse 4 or a second Bridge actually produce those sections. Essentially, the tags themselves haven’t changed, but the song-length limit increase and better coherence mean you can structure a song with many tagged sections (intro through outro) and expect v4.5 to follow through. (As a side note, the formerly used [extend-style] tag for continuing songs is less needed now, since the base model can generate extended compositions without a separate extend prompt.)
Improved Tag Prompt Parsing — Suno v4.5 introduced a smarter prompt parser, which affects how tags are interpreted within more natural sentences. Users have noticed that they can embed tags in a descriptive sentence and v4.5 still gets it — something v4.0 struggled with. For example: “The [chorus] should explode with [anthemic] harmonies and big drums.” In v4.0, that may have confused the model or caused it to sing the words, but v4.5 correctly reads those as tags (Chorus section; anthemic style). The outcome is that you don’t have to list tags stiffly on separate lines; you can mix them into a narrative prompt. The model’s better natural-language understanding means tags can carry more context. A Reddit user noted that steering the song with style tags “is a lot better now [in 4.5]” and you can use more natural phrasing around them. This update doesn’t introduce new tags per se, but it expands the way existing tags can be used, allowing for creative prompt-writing that still yields the desired structured result.
| Предыдущая глава |
↓ Содержание ↓
↑ Свернуть ↑
| Следующая глава |
