What lyrics to song AI actually does
A lyrics-to-song generator takes written words and produces a recording: a melody, a sung vocal, a backing arrangement and a structure. You supply the text and a short style direction. The model decides the tune, tempo and instruments.
That last point shapes how you should write. The model is not setting your words to a fixed tune, it is inventing a tune that fits the words. Lines that scan naturally give it an easy rhythm to follow. Lines of wildly different lengths, or paragraphs of prose, make it guess where the breaths and beats go.
TuneSplit sends your lyrics to two models, MiniMax Music v2 and ACE-Step, and returns two takes with different melodies and arrangements. If one stumbles on a line, the other helps show whether the fault is in the model or in the lyric.
How to format lyrics so the AI sings them well
Most disappointing results trace back to the text, not the model. The checklist below is what we would do before pressing the button.
Label every section on its own line
Write a tag in square brackets above each block: [Verse], [Chorus], [Bridge], and if you want them [Intro], [Pre-Chorus] or [Outro]. Lyric-to-song models are generally built to treat such labels as structure cues, which tells the model where the hook is and where to lift or drop the energy. We have not measured how each model weights them, so the practical test is to listen. If a label is sung aloud as a word, delete it and regenerate.
- Put the tag on a line by itself, with a blank line between sections.
- Use the same wording each time: always [Chorus], not [Chorus] then [Hook].
- Put the label above the words it applies to, never in the middle of a line.
- Number verses if you like ([Verse 1], [Verse 2]) so the order is clear.
Keep line length and syllables even
A sung line works best when it fits one breath. As a rule of thumb, aim for roughly 6 to 12 syllables a line, and keep lines within a section close to each other in length. If line one has 8 syllables and line two has 19, the singer has to cram the second one. Read each line aloud at a steady beat, and trim or split any line you cannot say in one breath.
Rhyme helps the melody land, but need not be exact. A pattern like AABB or ABAB gives the tune a place to resolve, and near-rhymes are fine in song.
Repeat the hook
Write the chorus once, then repeat the same text each time it returns instead of writing a new version. Repetition tells the model this is the hook. A short title line sung two or three times inside the chorus is the oldest trick in songwriting.
Write plain text only
Avoid stage directions, parentheses full of notes, emoji, long lists of ad-libs and anything that reads like an instruction. Use punctuation lightly: commas and full stops mark where a phrase ends. If you want a spoken or whispered moment, say so in the style box rather than in the lyric.
Lyrics length: how many words fit in a song
A song has a time limit, and so does your lyric. Free songs on TuneSplit run up to 2 minutes, Pro up to 3 minutes and Premier up to 4. A full song is made of a few sections, and each sung line takes a few seconds, so a lyric that is too long for the limit cannot all be sung at a natural pace.
| Length goal | Typical shape | Free plan fit |
|---|---|---|
| About 1 minute | One verse and one chorus | Fits easily |
| About 2 minutes | Verse, chorus, verse, chorus, short bridge or outro | Fits the free limit |
| 3 to 4 minutes | Two or three verses, repeated chorus, bridge | Needs a paid plan |
As a rule of thumb, a verse of four lines and a chorus of four lines each take well under half a minute. So two verses, two choruses and a short bridge is about right for 2 minutes, and a lyric with five long verses is not.
What happens to lyrics that are too long? The song is capped by the length limit, so the generator has to fit the text inside it. That can mean the vocal rushes, some lines are shortened, or the end of the lyric is left out. We cannot promise which, and it differs from model to model, so cut to fit before you generate. If you must keep a long lyric, split it into two songs: verses one and two in the first, the rest in the second.
A real example: lyrics in, two versions out
Here is a lyrics-to-song run, shown as input and output. We made Sunlight while building the site, so there is no copyright question. The style box said: Upbeat indie pop, bright female vocal, warm synths, claps, summer road trip feeling, 112 bpm. The lyric opened with Windows down and the radio low, Chasing the light on the coast road slow, and its chorus repeated the title: Sunlight, sunlight, carry us home.
Version A came from MiniMax Music v2: it runs 2:30 and took 94 seconds. That is longer than the 2-minute limit that applies to free songs. Version B came from ACE-Step: it runs 1:44 and took 27 seconds. Same words, two arrangements. Compare how each one treats the repeated chorus line. The short, repeated title is the part that sticks in both.
Notice what the lyric gives the models: short lines of similar length, concrete pictures, and one chorus line that comes back. Late Shift applies the same approach to a rap. The prompt was boom bap hip hop with a confident male rap vocal, and the lyric opened with Clocked in at nine when the city goes blue. MiniMax Music v2 made it in 59 seconds, 1:08 long.
Paper Boats is a gentler lyric: an acoustic folk ballad with a gentle male vocal, fingerpicked guitar and soft cello, from MiniMax Music v2, 1:30, made in 91 seconds.
Pick style, mood and vocal to match the words
The lyrics decide what is sung. The style box decides how. Match them: a tender lyric with short lines suits acoustic folk or soft R&B, a pile of internal rhymes suits rap or trap, and a long singalong chorus suits pop, country or rock. The style list is pop, lo-fi hip hop, acoustic folk, rock, R&B, EDM, trap, country, jazz, cinematic, k-pop and reggaeton, and you can also type free text.
| Setting | Options | Tip for lyrics |
|---|---|---|
| Style | Pop, lo-fi hip hop, acoustic folk, rock, R&B, EDM, trap, country, jazz, cinematic, k-pop, reggaeton, free text | Choose a style whose rhythm fits your line length |
| Mood | Happy, sad, romantic, energetic, chill, epic, funny, dark | Mood should agree with the words, or the song feels odd |
| Vocal | Female, male, instrumental | Instrumental ignores the lyric, so use it when you want only the music |
Do not name a living artist: TuneSplit has no sound-alike requests, voice cloning or covers in a real singer's voice. Describe instruments and feel instead, such as fingerpicked guitar or bright synths.
Language: lyrics in other languages
You can paste lyrics in the language you write in. Quality varies by language, and these models are generally strongest on the languages they saw most. If a line is mispronounced, the cure is usually to simplify the line, spell a tricky name the way it sounds, or put it in a separate short line.
Mixed-language lyrics are riskier, because the model must switch pronunciation mid-line. Keep each line in one language, and listen to both versions, since one model may handle your language better.
Edit and regenerate: the quick loop
Expect to run a song more than once. Each run costs 10 credits and returns two songs, so the free 30 credits a day give 3 generations. A failed generation is refunded automatically. Change one thing per run.
- Listen to both versions in full, with the lyrics on screen. Mark every line that sounded wrong.
- Fix the lyric first. Shorten the line, simplify the word or move the label. Text problems are the most common and the cheapest to fix.
- Then change the style if the sound is wrong. If the song is fine but the feel is off, change only the mood.
- Regenerate. The model starts over, so the melody will differ even with the same words.
- Keep the best take. Download the MP3 and note the settings that worked.
If you do not have lyrics yet, the lyrics generator writes them from a topic, style, mood and language. It gives 5 free drafts a day, and its Make this a song button sends the text here.
Who owns the song you make from your lyrics?
The words are yours. You wrote them, and uploading them to a generator does not change that. The recording is a separate question. Free songs on TuneSplit are for personal, non-commercial use. A commercial licence is part of the Pro plan at $9.90 a month or $99 a year.
Do not paste lyrics you do not have the right to use, such as another writer's song. Lyrics are moderated, there is a DMCA contact on the Terms page, and you should not ask for the sound of a particular artist. If you plan to release a song commercially, check the terms on the plan you use before you publish.
How TuneSplit compares with other lyrics-to-song tools
As of Oct 2026, aisongmaker.io has a dedicated lyrics-to-song page with a long guide and an FAQ, and it is a capable tool. We could not read a lyrics page for Suno, and several other lyric pages we tried did not load, so those cells say not verified.
| Tool (as of Oct 2026) | Free tier | Entry paid plan | Lyrics-specific guidance |
|---|---|---|---|
| TuneSplit | First song with no sign-up, then 30 credits a day (6 songs) | Pro $9.90 a month | Formatting, length limits and edit loop on this page; lyrics generator included |
| aisongmaker.io | 30 credits a day | Basic $14.99 a month | General guide on its lyrics page |
| aisong.io | 30 credits a day, 2 songs per generation | $14.99 a month | Not verified |
| Suno | About 50 credits a day | Pro about $10 a month | Not verified |
| MusicGPT | 500 credits a month (third-party figure) | About $9.99 a month (third-party figure) | Not verified |
Suno has a large community and library, and aisongmaker.io offers separate lyrics, text and upload modes. TuneSplit is newer and smaller, and has no upload-a-recording-to-song mode. Its edge is the work around the song: two versions per run, a lyrics generator, stems, a BPM and key finder and a lyric video maker.
After the song: video, stems and sharing
- Lyric video maker: paste the same lyrics, and the lines are timed across the song automatically. Export an MP4 in 9:16 or 16:9, rendered in your browser.
- Stem splitter: split your song into vocals, drums, bass and other. A 30-second preview of your own song is free, and full WAV stems cost 20 credits.
- BPM and key finder: check tempo and key before mixing with other audio.
- Share page: every song gets a public link and an MP3 download.
Free lyric videos carry a small Made with TuneSplit watermark, and Pro removes it.
Frequently asked questions
How do I turn lyrics into a song with AI?
Paste your lyrics into the TuneSplit song generator in lyrics mode, choose a style, a mood and a vocal, and generate. You get two versions from two models, a player, an MP3 download and a share link. No sign-up is needed. Adding [Verse] and [Chorus] labels helps the structure.
Is the lyrics to song generator free?
Yes, within limits. No sign-up is needed, and each browser gets 30 free credits every UTC day, which cover 3 generations, which is 6 songs. Free songs run up to 2 minutes and are for personal, non-commercial use. A failed generation is refunded automatically.
What format should my lyrics be in?
Plain text with a label such as [Verse], [Chorus] or [Bridge] on its own line above each section, and a blank line between sections. Keep lines in a section similar in length, repeat the chorus text exactly, and leave out stage directions or emoji. Then listen and adjust.
How long can my lyrics be?
Long enough to fit the song length: free songs run up to 2 minutes, Pro up to 3 and Premier up to 4. About two verses, two choruses and a short bridge suit 2 minutes. Beyond that the song cannot sing everything at a natural pace.
What happens if my lyrics are too long?
The song is capped by the length limit, so the model must fit the text inside it. That can mean a rushed vocal, shortened lines or a missing ending, and we cannot predict which. Trim the lyric before generating, or split a long lyric into two songs.
Do the [Verse] and [Chorus] tags get sung?
They are meant as structure cues, not words to sing. We have not measured how each model treats them, so the practical test is to listen. If a tag is sung aloud, delete it or reword it and regenerate. Putting each tag alone on its own line gives the best chance.
Can I use lyrics in another language?
Yes, paste lyrics in the language you write in. Quality varies by language, and mispronounced words are the usual issue. Simplify the line, respell a tricky name phonetically, or keep each line in one language. Listen to both versions, as one model may handle your language better.
Can I use the song commercially?
Not on the free plan, which is for personal, non-commercial use. A commercial licence is part of the Pro plan. The lyrics you wrote remain yours either way. Check the terms of your plan before publishing.
Why does the song not sing my lyrics exactly?
The models write a melody to fit the words, and sometimes shorten, repeat or restyle a line, especially when lines are very long or uneven. Shorten the line, even out the syllables, and regenerate. Because two models run at once, compare both versions before changing the text.
How long does it take to make the song?
The ACE-Step version usually arrives in 10 to 30 seconds and the MiniMax Music v2 version in about 60 to 95 seconds. The first playable version usually comes within a minute. Our measured examples ranged from 11 to 94 seconds, but busy periods can add time.
Can I change just one line and regenerate?
You can edit any line in the lyrics box and run the generator again, which costs 10 credits for two new songs. The model starts over, so the melody changes even with unchanged lines. Change one thing per run, such as a line or the mood, so you can tell what helped.
Are my lyrics moderated or made public?
Lyrics are moderated. Each song has a public share page, so do not paste anything private into the box. Only use lyrics you wrote or have the right to use. Rights holders can send a DMCA request through the contact on the Terms page.