|
↓ Содержание ↓
↑ Свернуть ↑
| Следующая глава |
Suno meta-tags
Last modified: 2025, August 26.
Suno service news (updated: July 18, 2025)
On 18-th of July Suno has released for paid subscriptions, access to a new model, "v4.5+"
The initial tests showed that v4.5+ generated more variable, richer tracks with better polyphony and overall quality. Typical restrictions are the same as for v4.5. It's not yet known if there are changes in how v4.5+ interprets meta-tags.
Also, the "Inspo" tool has been added to tracks generation UI. That, according to the UI hint, uses songs from specified playlist as "source for inspiration" In practice, Suno blends the "inspiring" tracks, making kind of pot pourri in the output. No formal definitions for "Inspo" was officially done yet.
Riffusion relation
Since April 27, 2025 Riffusion.com audio tracks generation service (very much like Suno) introduced paid levels of membership and a new model FUZZ-1.0 Pro.
After 400+ generations that were first tested in Suno and then in Riffusion ("Style of Music" at Suno going to "Sound" prompt at Riffusion; "Lyrics" going to "Lyrics") it was discovered that Riffusion generally follows the same set of official and user-tested meta-tags known for Suno.
Known misbehaving tags/tags values (consider obsolete)
Tag
Status
Notes
[autotune: ...]
❌ Unsupported
[filter: ...]
❌ Inefficient
[loop: ...]
❌ Unsupported
[mix, mixing: ...]
❌ Inefficient
[master: ...]
❌ Inefficient
[pan, panning: ...]
❌ Inefficient
[style: none]
❌ Invalid
"none" is not interpreted as a meaningful style. Will confuse output.
[section: ...]
❌ Redundant
Rejected or misread; use [intro: ...], [verse: ...] etc.
[theme: ...]
❌ Invalid
unlabelled theme tag is ignored or causes parsing errors, use [theme A: ...] etc.
[volume: ...]
❌ Inefficient
Tips on using meta-tags
Use only known or confirmed tags.
Avoid alias or ambiguous tags like bpm, key, language.
Test in Standalone mode first — many tags that break in Extend or Cover work fine solo.
If in doubt, use "[style: experimental]" or "[control: hallucinatory]" to encourage flexible output instead of forcing
Parsing Note for Suno v4.0+
Any unrecognized tag (e.g. [emotion build], [tense development]) is treated as a generic [section X: ...] tag.
Use colon syntax to embed directives safely:
[tense development: Style and mood of the theme gets more tense until a climactic counterpoint is reached]
Avoid placing instruction text outside the tag, as it will be sung/spoken:
[tense development]
Style and mood... ← ❌ becomes lyrics
Avoid inserting lyrics/text to sing as [verse] colon notation: the engine will only sing that text if doesn't find obvious rendering directives; place the verse lyrics immediately below its [verse] tag
To prevent looping: do not repeat identical lyrics or tag sections verbatim; add variation or split across multiple [verse A], [verse B] etc.
For structure: always include [sequence: ...] to control order when extending or building long-form compositions.
Version-Dependent Differences in Tag Behavior (v4.0 vs v4.5)
Suno v4.0 and v4.5 share the same meta-tag system in spirit, but there are notable differences in how tags are interpreted and how reliably they influence the music between these two versions. Key version-dependent differences include:
Natural Language Prompting: Suno v4.5 is significantly better at understanding descriptive, sentence-form prompts with embedded tags, whereas v4.0 tended to respond mostly to more rigid, shorthand tags. In v4.0, users often listed a sequence of tags or keywords (e.g. [Intro], [Verse], [Chorus], or Mood: energetic | Genre: Pop | Vocals: Female etc.) and kept instructions terse. With v4.5, the model can parse richer language around tags. The Suno v4.5 update was tuned to "understand and translate your descriptions" of mood, instruments, and style with more nuance. An official guide notes that v4.5 still supports the old bracket tags format, "but now responds better when those tags are embedded into natural, full-sentence instructions.". In short, v4.0 required more prompt "keyword telegraphing," while v4.5 allows a more conversational prompt with tags inside it. For example: instead of just writing [Chorus: anthemic], a v4.5 user might write a full line like "The [chorus] should hit with [anthemic] vocal harmonies and huge drums," and v4.5 will follow that more accurately than v4.0 would. This reflects improved prompt adherence in v4.5. Users have found they can be more verbose and specific in v4.5 without confusing the model.
Prompt Adherence and Reliability: Overall v4.5 is more reliable in obeying meta-tags than v4.0 was. Suno v4 sometimes ignored or only loosely followed tags (especially complex style directives), leading to user frustration. In fact, some early v4.0 users complained that "I can’t get Suno to obey the style tags, it’s as if they are not used at all...". By contrast, v4.5’s improved prompt fidelity means tags like [Intro], [Bridge], [Mood: X], etc., more consistently shape the output. A Redditor noted that "steering the final product with style tags is a lot better now [in 4.5]" and that they can even use more natural phrasing instead of hunting for the perfect v4-era keyword. In summary, where v4.0 might gloss over a bracket tag or require multiple re-rolls to get the effect, v4.5 more often gets it right on the first try (especially for genre and mood tags, thanks to prompt adherence improvements).
Structural Tags and Song Length: Both versions support tags like [Verse], [Chorus], etc., but v4.5 can handle longer structures. Suno v4.0 was limited to about ~4 minutes of song, often requiring an Extend feature to continue. In v4.5 the max length doubled to 8 minutes, and the model maintains coherence over that length. This means tags delineating multiple verses and choruses (or multiple [Theme A], [Theme B] sections in a long piece) are more feasible in v4.5. The tags themselves didn’t change, but the behavior is that v4.0 sometimes truncated or ignored later structural tags if the song was too long, whereas v4.5 can actually execute a full structure from intro to extended outro with all tags accounted for. Additionally, v4.5 introduced an improved "Cover+Persona" mode and better Extend, but those are features beyond the lyric meta-tags (except that [extend-style] tag from the doc is less needed now since v4.5 can natively go long).
Deprecated vs Replacement Tags: The transition from earlier versions to v4 introduced some tag deprecations. For example, as noted above, [sing-style] was used in Suno’s early alpha but by v4.0 it was replaced by [vocal-style]. Similarly, [song-type] was an early tag to indicate if the piece should be a song, rap, instrumental, etc., but by v4.0 it was rendered inert. Another subtle change: older prompts sometimes used [style] for genre/style, but in v4.0+ it became more effective to use [genre:] and specific style qualifiers (because "[style: none]" had proven invalid and you needed to actually name a style). These differences were captured in the documentation’s errata, but it’s worth noting they specifically affect v3-era vs v4-era usage. By the time of v4.0, those older tags were already phased out, and v4.5 continued to ignore them. In practical terms, a user coming from Suno v3.5 to v4.0 had to unlearn a tag like [bpm] or [section]. But between v4.0 and v4.5, there were not many new deprecations — mostly improvements in understanding rather than removal of tags (the major tag removals happened earlier, in the jump to v4).
Parameter Interpretation: Some tags retained the same name but improved their parameter handling in v4.5. For instance, the [tempo:] tag in v4.0 could accept broad terms like "slow" or "fast." Suno v4.5 still doesn’t take exact BPM numbers, but it got smarter about tempo descriptors — it will pick up on more nuanced phrases like "mid-tempo 90s hip-hop swing" within a tag or prompt. Another example is the [vocal-style:] or [vocal-tone:] parameters: v4.0 might understand basic values (whispered, raspy, etc.), but v4.5 has a greater range of vocal texture it recognizes (you can specify things like "nasal, twangy" or "smooth crooning" in v4.5 and it more likely yields a difference, whereas v4.0 would often ignore such fine detail unless it was a preset tag). The underlying model upgrade in v4.5 "captures subtle musical elements" better, which extends to tag parameters for subtle dynamics (like the difference between a "soft" and "intense" whisper in [whisper:]). The documentation’s tag definitions didn’t change, but the outcome of using certain parameters (like [chorus: soft] vs [chorus: anthemic]) is more distinct in v4.5 than it was in v4.0, thanks to the model’s improved fidelity.
Genre Tag Expansion: As mentioned, v4.5 expanded genre support. This isn’t a syntax change but a behavior change: a tag like [genre: jazz-house] or [genre: midwest emo] might have confused v4.0 (or defaulted to one genre, ignoring the hybrid), whereas v4.5 handles multi-genre combinations much more gracefully. The result is that some meta-tags that v4.0 would effectively not honor suddenly became meaningful in v4.5. For example, if you put [genre: punk rock meets classical] in v4.0, you’d likely get something incoherent or just one genre dominating. In v4.5, the same tag prompt can actually yield a convincing punk/classical crossover because the model learned genre blending. Therefore, users in v4.5 can utilize more imaginative genre tags or mashups, which is a new capability not reflected in the older documentation of tags.
In essence, Suno v4.0 and v4.5 use the same set of meta-tags for the most part, but v4.5 interprets them more accurately and with a broader palette. V4.5 encourages more descriptive usage of tags (embedding them in sentences, stacking multiple attributes) whereas v4.0 required a more minimal, list-based approach. This means certain tags that were technically available in v4.0 only truly became useful in v4.5. A concrete example: [Mood: Uplifting] [Genre: Gospel] [Style: Lo-fi] might yield something muddled in v4.0, but in v4.5 one could write, "Create an uplifting lo-fi gospel piece — [Mood: uplifting] [Genre: gospel] [Style: lo-fi], with a choir and dusty vinyl crackle," and it will surprisingly adhere to that vision. The tag names didn’t change, but the behavior and fidelity did from v4.0 to v4.5. Users should note these improvements when crafting prompts for the respective versions.
Instrumental vs. Song (Vocal) Tag Usage
Suno’s meta-tag system covers both purely instrumental music and songs with vocals, and some tags are more relevant to one or the other. It’s important to distinguish how tags apply in instrumental tracks versus vocal tracks, and note any differences in v4.0/v4.5 behavior for each category:
Instrumental Tracks: To generate an instrumental (no vocals) piece, the key tag is [instrumental]. This tag explicitly tells Suno not to produce vocals, focusing on instruments only. The documentation defines [instrumental] as "ensures the track contains no vocals". In both v4.0 and v4.5, placing [instrumental] at the start of the lyrics prompt is the recommended way to get a music bed with zero singing. V4.5 seems to honor this even better (v4.0 sometimes would slip a faint vocal hum or oh’s, but v4.5 is more strict about it, likely due to improved prompt adherence). When using [instrumental], you can also specify a style parameter (e.g. [instrumental: orchestral cinematic composition]) to guide the flavor.
For instrumental pieces, structure tags like [Intro], [Verse], [Chorus] still can apply — they will just denote purely instrumental sections. For example, you might do:
[instrumental]
[intro: Slowly building strings and piano]
[verse: Main melody introduced on guitar]
[chorus: Full band enters with drums and bass]
This is valid in both v4.0 and v4.5. The difference is, v4.5 will likely produce a more coherent instrumental "song" with those sections (taking care to change up the instrumentation per tag), whereas v4.0 might have been more repetitive without vocals to lead. The documentation even gave track structure recommendations for instrumentals using normal section tags (intro, verse, etc.) but no vocals, illustrating that structural tags are not exclusively for sung lyrics.
Additionally, instrumental tracks often use solo and instrument tags extensively. Tags like [Instrument: Piano] or [Guitar Solo] become the "lead voice" in absence of vocals. Suno v4.5 can follow these well — e.g. [Instrument: Violin (Lead)] would likely make violin carry the melody. In v4.0 it also works, but perhaps with less nuance. The user doc defines a generic [instrument] tag to highlight a particular instrument in the arrangement, and an [instruments] tag to list multiple instruments for the whole track setup. Both of these are quite useful for instrumentals. For instance, one could prompt: [instruments: acoustic guitar, cajón, handclaps] to set the timbre palette of an instrumental. Suno v4.0 and v4.5 both pay attention to these lists, though v4.5 is better at correctly blending unusual combinations.
It’s also worth noting the [solo] tag (or instrument-specific solos) are primarily instrumental in nature. The doc provides [solo: ...] as a tag meaning an improvised instrumental solo section. In a song with vocals, a [solo] usually means an instrumental break (guitar solo, etc.). In a purely instrumental track, a [solo] might simply mean a single instrument is spotlighted. Both versions support it, but again v4.5 tends to produce more convincing solos (e.g. an electric guitar solo that actually sounds distinct and lead-like, whereas v4.0 solos might sound more like a continuation of the backing track). Community tags like [Drum Solo] or [Instrumental Break] are effectively specialized cases and work similarly across versions.
Songs with Vocals: For tracks that include singing or rapping, there are many vocal-related tags which wouldn’t apply to instrumentals. Basic structural tags like [Verse], [Chorus], [Bridge], [Outro] are chiefly used in vocal songs to organize lyrics and musical sections. Suno v4.0 and v4.5 both rely on these to know where to generate verses vs choruses. Typically, verses have new lyrics, choruses repeat the hook. If you use these tags in an instrumental track, the model might still create contrasting sections instrumentally (for example, a "chorus" section could bring in a fuller arrangement even if no words). But their primary purpose is for lyrical structure. V4.5 showed improvement in handling these — e.g. it produces more dynamic, distinct choruses and bridges than v4.0 did. The user comparison of v4 vs v4.5 noted "more varied sections and transitions" in v4.5’s instrumental output as well, meaning it followed the intended structure better.
|
↓ Содержание ↓
↑ Свернуть ↑
| Следующая глава |
