Udio Lyrics Format: The Complete Section + Directive Reference
Udio does not publish a formal lyric spec, so most formatting advice for it is folklore. This reference takes the opposite approach: we maintain a provider layer that ships the same lyric to multiple music engines, and this is the portable bracket-tag format it emits — the structure vocabulary that transfers to Udio, the directive syntax that needs translating, and the mistakes that get your stage directions sung out loud.
Two kinds of brackets, two different rules
Everything in square brackets in a formatted lyric is an instruction, not a word to sing, and there are exactly two kinds. Section tags name the structure — [Verse 1], [Chorus], [Bridge] — and are the closest thing AI music has to a common standard: they communicate structure to every engine our provider layer has shipped against. Performance directives — [WHISPER], [FULL BAND DETONATION.] — are production instructions, and they are a convention, not a standard: some engines honor them, some ignore them, and some sing them as words.
That asymmetry is the whole key to formatting for Udio. Lean hard on section tags; treat directives as the fragile layer that needs testing and a graceful fallback. Each gets its own rules below.
The 14 section tags
Our format layer treats fourteen section names as canonical, and they cover essentially every block a modern song uses:
[Intro] · [Verse 1] · [Pre Chorus] · [Chorus] · [Post Chorus] · [Hook] · [Bridge] · [Break] · [Build Up] · [Interlude] · [Transition] · [Solo] · [Inst] · [Outro]
Three formatting rules, all load-bearing:
- One tag per line, nothing else on the line. A tag buried mid-line stops being a tag and becomes a lyric. Our normalizer only treats a bracket as structural when it is the entire line — a safe assumption to hold for any engine's parser.
- Numbers go inside the bracket, after the name.
[Verse 1],[Verse 2],[Pre Chorus 2]. Numbering matched sections helps an engine map verse 2 onto verse 1's melody. - Pick one spelling and hold it. Our layer folds
[Pre Chorus]/[Pre-Chorus],[Build Up]/[Build-Up], and[Inst]/[Instrumental]to one canonical form each, because a lyric that alternates spellings reads as two different sections to a less forgiving parser. When formatting by hand, just pick one form and use it everywhere.
On casing: our provider layer emits title case ([Verse 1]) when normalizing for external engines, and all-caps internally — both are widely understood, but title case is the conservative choice for Udio-bound lyrics. An invented section name outside this list usually won't break anything; it just gambles on interpretation instead of using vocabulary every engine has seen.
Performance directives: parentheticals degrade gracefully
Directives shape production — instrumentation shifts, dynamics, silence, vocal delivery, re-entry impact. In the all-caps bracket dialect they look like [CLEAN ELECTRIC GUITAR. ONE STRUM PER BAR.] or [DEAD SILENCE. NO BREATH.]. The problem: an engine that doesn't recognize a bracket as a directive may treat it as singable text, and a sung stage direction costs the whole take.
Our provider layer's translation rule is the one to copy. When shipping to engines that don't read the all-caps bracket convention, it converts directives to parenthetical directions: [FULL BAND DETONATION.] becomes (full band detonation). Parentheticals degrade gracefully — an ignored parenthetical costs nothing, while a sung bracket ruins the render. For Udio, the practical protocol is:
- Keep section tags in brackets — they're the standard layer.
- Write performance intent as lowercase parentheticals at the start of the section they govern:
(whisper),(drums drop out),(building, each line louder). - Put the big production decisions in the prompt box instead of the lyric — the prompt is the layer every engine definitely reads.
And keep the economy rule from the Suno tag reference: directives lose all meaning when every section has one. One per section is a sensible ceiling; a well-placed silence before the line that matters most is worth more than ten delivery cues.
Why clean formatting matters more on Udio
In our experience shipping the same songs to multiple engines, Udio weights the lyric itself more heavily than Suno does — Suno privileges the style prompt; Udio privileges the words. (The full comparison is in Udio vs Suno: which takes lyric direction better?) That cuts both ways: a disciplined lyric is preserved more faithfully, and a formatting mistake is amplified rather than papered over.
So the portable lyric disciplines pay double on Udio: matched sections with the same line count and syllable shape, rhymes that work out loud rather than on the page, open-vowel endings on the lines a melody will peak on, and the hook word in the strong position. None of that is Udio-specific — the singability rules are about the human voice, not any vendor's parser — but Udio is the platform where the lyric layer most directly becomes the record.
The prompt box: artist names and front-loading
Two rules from our format layer that hold on every platform we've integrated:
- No real artist names. "In the style of [famous artist]" is the one prompt pattern that fails across the board — every major engine we've shipped against refuses or filters real-artist style requests, and a single name can get an otherwise-clean generation rejected. Our pipeline scrubs "in the style of…" and "sounds like…" constructions from every prompt before it reaches any provider. Encode the intent as era plus production instead: "late-90s alt-country, pedal steel, close-mic'd female vocal" transfers; a name does not.
- Front-load the load-bearing information. Prompt length budgets vary by engine, so build the prompt as an ordered hierarchy — genre, vocal, core instrumentation first; atmosphere and detail last — and truncation costs you adjectives instead of the genre. Declaring the vocal gender explicitly is the cheapest insurance in the whole workflow; our layer literally prepends it when a prompt doesn't state it, because an unstated vocal is a coin flip on every platform.
The anatomy of a strong prompt is its own study — the style-prompt guide applies almost unchanged.
The four mistakes that garble a render
The failure modes we see most when a lyric moves to Udio, in order:
- Stage directions written as lyric lines. "The music swells as she remembers" outside any bracket will be sung, verbatim, in full sincerity — and on a lyric-forward engine it will be sung prominently.
- Mid-line or lowercase tags.
my heart (whisper) breaksis a lyric containing the word "whisper" in costume. Structure tags live on their own line; parenthetical directives sit at the section start, not inside a sung phrase. - Unmatched section shapes. A six-line verse 1 and a ten-line verse 2 force the engine to either rush the second verse or abandon the melody. Match line counts and syllable shape across same-named sections.
- A lyric that argues with the prompt. A grief lyric under an upbeat dance prompt produces the uncanny valley on any engine — and on a lyric-weighted engine, the tug-of-war is audible. Decide which document owns the mood and make the other one agree. (What transfers and what breaks covers the full porting checklist.)
A formatted skeleton to copy
The Udio-safe shape: bracketed section tags on their own lines, parenthetical directives at section starts, sung text everywhere else:
[Intro]
(single piano, no drums)[Verse 1]
(close, quiet vocal)
First lyric line here…[Pre Chorus]
(building, drums enter)
Pre-chorus lines here…[Chorus]
(full band)
Chorus lyrics here…[Bridge]
(strip back to voice and bass)
Bridge lyrics here…[Outro]
(fading instrumental)
The format carries the structure; the words still have to be worth singing. Our forge emits this whole package — formatted lyric plus a prompt built on the portable rules above — on every song, and you can forge one or paste your own draft into the Crucible for a free critique before you spend a render on it.