Overview

Suno uses Chirp to generate songs, which is a generative model. We must create documents that prompt the model correctly. To do this, we use specific tags to break the songs up into sections that the model understands.

Base Tags

This list represents all of the valid base tags. Each of these tags can be modified and has several ways they can be used, but in general these are the only valid tags

• [Intro]
• [Hook]
• [Pre-Chorus]
• [Chorus]
• [Verse]
• [Interlude]
• [Break]
• [Movement]
• [Instrumental]
• [Solo]
• [Build]
• [Bridge]
• [Outro]
• [End]
• [<vocals>] (clarification in comments)
• [<specific instrument>] (clarification in comments)

• [Intro]

The intro tag should generally only be used at the beginning and is strictly instrumental. It can be modified with several adjectives. For instance:

• [Long Mellow Intro]
• [Short Exciting Intro]
• [Dreamy Slow Intro]

As you can see from the examples, you can generally add an emotive and/or a pacing adjective. The system doesn't always honor the intention, but it tends to work best if you use very direct, concrete adjectives that are salient to musical construction (speed, emotion, intensity seem to work best). Modifiers are not strictly necessary, but can be useful for establishing the mood early on.

• [Hook]

Generally not necessary unless you modify it, more or less treated like an intro. Can be used to transition from intro to main part of song, particularly if the intro is different.

• Pre-Chorus]

This is a strictly vocal tag that is often used at the beginning of songs to sort of introduce the story or narrative. It may or may not be sung (could be spoken, can be specified). This tag should generally only be used once or before chorus tags.

• [Haunting Whispered Pre-Chorus]
• [Staticky Spoken Pre-Chorus]
• [Primal Scream Pre-Chorus]
• [Opera Female Pre-Chorus]

These modifiers, while not strictly required, can confer very specific feels. As with intro, emotive, intensity, and pacing adjectives tend to work well, with the added option of singing styles, gender, and so on.

• [Chorus]

This is about what you'd expect for any song construction. Generally speaking, the Chirp system decides how to render the chorus, so modifiers are often NOT honored, for whatever reason. The documentation says that the chosen style and lyrics tend to do more to modify how the chorus is sung. However, very concrete modifiers are most likely to be honored. This tag is one of the prime workhorses.

• [Whispered Chorus]
• [Eerie Chorus]
• [Ensemble Chorus]
• [Slow Chorus]

Lyric construction also has a huge impact on how the chorus is sung, and is often more important once the construction has been set up correctly. The system seems to ignore capitalization, but the vibe of the lyrics has an impact. Punctuation seems to have a larger impact

- Elipsis... this tends to make the system approach it more slowly, particularly if it's used... multiple times... in the line...
- Exclamation! this doensn't often have a huge impact but it can tell the system to emphasize a line
- Oooooohhh whoaaa ahhhh! vocalizations generally work extremely well to amp up a chorus (the system will not render non-word vocalizations without explicitly being told)
- mmmmmmmmmmm oh... gentler vocalizations will have a similar dampening effect
- (parenthesis lyrics) this does really well, and seems to do really good at making it do call-and-response or antiphonal

• [Verse]

This tag is the other primary workhorse and is used pretty much identically to [Chorus]. It is not strictly necessary to modify it, and in many cases the system will decide how to modify it (consider that in most of it's training data, chorus and verse are left plain). So again, the content of the lyrics tend to do more. The system seems particularly sensitive to where in the song it is, meaning that if the generator feels like it's more likely a crescendo rather than a key change, the system will make that judgment call. However, emotive, intensity, and pace modifiers tend to work.

• [Angry Verse]
• [Mysterious Verse]
• [Whispered Verse]
• [Spoken Verse]
• [Opera Verse]

These concrete modifiers are most likely to be honored. The same lyrical modifiers from the chorus also apply

- Elipsis...
- Exclamation!
- Vocalizations
- (Parenthesis)

For example:

I am the void between stars
(Beyond the veil of forms...)
I am the death of light
(Where your deepest terrors remain...)

• [Interlude]

This is one of the main workhorses of the instrumental tags. It's pretty much what you'd expect. Many modifiers don't seem to impact this tag, but a few tend to be more reliable.

• [Melodic Interlude] this one is pretty reliable, so similar modifiers should work.
• [Long Melancholy Interlude] this works about half the time
• [Short Accelerating Interlude] the system tends to prefer short instrumentals anyways

Don't get too creative with the modifier tags. For instance, genre-specific modifiers don't really seem to work like `[Psychedelic Interlude]` even though it makes sense to us, the system doesn't seem to recognize it. However, there is another option we have to modify all instrumental sections, and that is periods and exclamations to try and shape the pacing.

[Melodic Interlude]
. . . ! . .
. ! . . . !

[Intense Interlude]
!! . ! !! !
!! !! ! !!

And so on. You can arrange the . and ! in any way you like to convey the rhythm and such.

• [Break]

Break is strictly instrumental often defaults to one measure or phrase, and can be used quite frequently. Almost no modifiers work here, and it has the most impact on the song when wedged between verses and choruses. What does tend to work is specifying the instrument to lead during the break:

• [Violin Break]
• [Drum Break]
• [Scream Break]
• [Lead Guitar Break]
• [Bass Guitar Break]

So basically you can use it like a small solo. The rhythm modifiers are generally totally ignored here.

• [Movement]

This is an experimental tag, but _might_ help the engine transition to a new movement.

• [Begin Psychedelic Movement]
• [Transition to Faster Harder Movement]
• [Long Orchestral Movement]

The system is liable to totally ignore this tag but it's worth a shot.

• [Instrumental]

General purpose tag to break up a song. Can be used on its own, unmodified. Often used in conjunction with other tags or modifiers (see rest of doc)

• [Solo]

This tag is pretty much exactly what you'd expect. It pairs well with [Interlude] and does best when you specify the instrument, pace, and energy.

• [Soaring Lead Guitar Solo]
• [Fast and Intense Drum Solo]
• [Dancing Fiddle Solo]
• [Playful Flute Solo]
• [Finger Style Guitar Solo]

As you can see, this kind of pattern tends to do best with the solo. It all generally comes back to instrument, pace, emotion/energy. One thing to note is that chaining [Interlude] and [Solo] is often the best way to change the movement or overall tone of a song. The exclamation and period modifiers sometimes work on solos, but the system often just goes with the vibe.

• [Build]

This tag is often less effective than interlude or solo, and is often treated like a break. Period and exclamation rhythms seem to have relatively little effect, though it seems to have a decent effect when sandwiched between verse and chorus (e.g. to build to a soaring chorus) but that can be redundant. This should only be used when it really makes sense for the song, and probably only once. In most cases, a break, solo, or interlude should be used, that's how specific this use case is.

• [Bridge]

The system doesn't seem to know what to do with bridges. It often just treats them like a verse or a chorus, sometimes a refrain. Use them sparingly, at most once per song.

• [Instrumental Bridge] - this seems to have the most impact or be most useful

• [Outro]

This tag seems to work if it's either instrumental or vocal, and can be treated pretty flexibly. It's primary purpose is to tell the system to start preparing for the end of the song and should be used exactly once.

• [Long Fading Outro]
• [Urgent Loud Outro]
• [Mournful Outro]

Like many such tags, emotion + pace seems to work well. This tag should only be used once near the end to cue the generator to start winding the song down.

• [End]

As you'd expect, this generally tells the system to end the song.

• [Fade to End]
• [Lingering End]
• [End Resolves to Whispers]

You can play with this tag some, but generally it just serves as a standalone tag. Some of the same modifiers here can work for the outro tag as well.

More in the comments...