You wrote 200 words of lyrics and Suno sang something else entirely — not because the model is “improvising”, but because it treated your text as a theme description rather than lyrics. Suno’s Custom Mode follows your text word-for-word; Simple Mode treats it as a prompt and writes its own lyrics. 99% of “lyrics ignored” reports are the wrong mode or missing structure tags.
To get Suno to sing exactly what you wrote, use Custom Mode + structure tags + lyric-to-length matching.
Common causes
By how often each one drops the lyrics:
1. Simple Mode instead of Custom Mode
Two modes:
- Simple Mode: you give a description (“a happy song about coffee”), AI writes lyrics and music
- Custom Mode: you provide lyrics + style, AI sings your lyrics verbatim
Pasting lyrics into the Simple Mode prompt box → model uses them as theme, rewrites everything.
How to judge: is the “Custom Mode” toggle on at the top of the UI? If off, you’re in Simple Mode.
2. No [Verse] / [Chorus] structure tags
Even in Custom Mode, plain-text lyrics without tags let the model segment lyrics its own way. Common result: the chorus you wrote gets sung as a verse, your verse becomes a bridge — lines get reordered.
How to judge: do your lyrics include [Verse] / [Chorus] / [Bridge] style bracket tags? If no, this is it.
3. Lyric count doesn’t match song length
Suno defaults to 2-3 minute songs. Rough line budget:
| Length | Line count (English / Chinese) |
|---|---|
| 1 min | 12-16 / 8-12 |
| 2 min | 24-32 / 18-24 |
| 3 min | 36-48 / 28-36 |
| 4 min | 50-65 / 40-50 |
100 lines in a 2-minute song → model has to pick a subset, the rest get dropped.
How to judge: lines/minute > 32 (EN) or > 25 (ZH) means too many.
4. Style describes a vocal-less arrangement
instrumental, background music, lo-fi beats hint “no vocals or sparse vocals” — the model might hum a few lines and skip the rest.
How to judge: does style contain instrumental / no vocals / background music?
5. Special characters / emoji in lyrics
💔, <3, emoji, special markdown — these get filtered as non-text by the model.
6. Using a Persona whose vocal style doesn’t fit your lyrics’ meter
Some Personas are rap-trained but you wrote a ballad — phrasing gets messed up and the lyrics sound mangled.
Shortest path to fix
By lyric-hit rate:
Step 1: Switch to Custom Mode
Top-right of the UI, toggle “Custom Mode” on. You now get two separate fields:
- Lyrics (left)
- Style of Music (right)
Lyrics belong only in the Lyrics field — never paste them into Style.
Step 2: Add structure tags
Rewrite lyrics as:
[Verse 1]
I opened that old album
You were laughing on the first page
Outside the rain fell and stopped
Like the promise we never finished
[Pre-chorus]
Late at night I'm still awake
[Chorus]
Turns out you'd already left
What's left is just the echo
Turns out you'd already left
And I'm learning to be alone
[Verse 2]
...
[Bridge]
...
[Chorus]
...
[Outro]
Turns out you'd already left
Tag list (by frequency):
[Intro],[Outro][Verse 1],[Verse 2][Pre-chorus],[Chorus][Bridge],[Hook][Instrumental](no vocals in this section)[Spoken](talk, don’t sing)
Step 3: Match line count to song length
Trim or extend to budget:
- 2-min song: 24-32 EN lines
- 3-min song: 36-48 EN lines
50+ lines? Split into two parts (part 1 / part 2) or stretch to 4 minutes (v4 supports up to 8).
Step 4: Don’t write vocal-less styles
Avoid:
instrumental, background music, lo-fi beats
Write:
85 BPM, indie pop, soft female vocal, melancholic, fingerpicked guitar
Include vocal explicitly so the model knows lyrics need singing.
Step 5: Strip special characters
Remove from lyrics:
- All emoji (💔, 🎵)
- Markdown symbols (bold, italic)
- HTML (
<br>,&) - Excess blank lines (don’t leave 3+ empty lines)
Keep: punctuation, single-line breaks, [Section Tags].
Step 6: Still skipped? Force with Replace Section
If a specific section (e.g. bridge) keeps getting skipped:
- Generate, then select that section
- Replace Section
- In the new input, only the lyrics for that section with an explicit
[Bridge]tag
Prevention
- Always use Custom Mode; Simple Mode is for “let me hear something random”
- Add
[Verse]/[Chorus]tags even on short tracks - Match line count to length: 2 min ≈ 28 EN lines, 3 min ≈ 42 EN lines
- Style must include
vocalor it may default to instrumental - Strip all emoji / markdown / HTML from lyrics
Related
Tags: #Suno #Music #Debug #Troubleshooting