Why Suno ignored your Style prompt

You asked for lo-fi bedroom pop and got a stadium chorus with a polished mix. You add more words to the Style field, generate again, and it drifts somewhere else. At some point you conclude the field is decorative.

It is not being ignored. It is being averaged, and an average of contradictions is nobody's song.

Sound goes in the Style field, and only sound

The first thing to rule out is instructions in the wrong box. Suno performs what is in the lyrics field, so a note to the singer becomes a line the singer sings:

[Verse 1]
(slow and sad, 80 bpm)
I drive back to the place

The parenthetical is not a setting. Move it to where settings live and the verse starts on the first real line:

[Verse 1]
I drive back to the place

Tempo, mood, instruments, production and voice all belong in the Style field. The free Suno lyrics check reads both boxes and flags the ones that crossed over, along with the Style field itself, which is the harder half.

Eleven descriptors, four of them arguing

Here is a Style field of the kind people actually paste, written for this piece:

warm and intimate, lo-fi bedroom pop, stadium chorus, acoustic guitar,
808s, dreamy, melancholic, upbeat, polished radio-ready mix,
female vocals, 140 bpm

Nothing in it is unreasonable on its own. Run it through the check and five findings come back, in the words the tool prints:

Fix       "warm and intimate" and "stadium chorus" pull in opposite directions. Keep one.
Fix       "acoustic guitar" and "808s" pull in opposite directions. Keep one.
Fix       "lo-fi bedroom pop" and "polished radio-ready mix" pull in opposite directions. Keep one.
Fix       "upbeat" and "melancholic" pull in opposite directions. Keep one.
Consider  "warm and intimate" comes before the genre (lo-fi bedroom pop).
          Put the genre first so everything else is read in its light.

Each finding carries a button, "Fix the tags in Heho", and a line under the list: "Free account. Your lyric comes with you into the editor." The rewrite at the end of this piece took no account and about a minute.

Four pairs, each of which a human would have to choose between. The model does not get to refuse the brief, so it splits the difference, and the difference between intimate and stadium is a room nobody recorded in.

The tag-count argument is not settled, and it is the wrong argument

The community disagrees about this, loudly. The best-liked reply on a long style guide puts one side plainly: "if I want a thrash metal song I type in 'Thrash Metal' and that's it. I have no need or interest in prompts that look like Einstein's Theory of Relativity."

The other side, on the subreddit's meta tag guide, argues the opposite with equal confidence: write as much as you can everywhere, describe rather than list keywords, because there is a language model in between that makes sense of whatever you give it.

Both are describing real results, which is the tell that the count is not what matters. The example above has 11 descriptors and never trips the check's ceiling of 12. It still contradicts itself four times. A one-word Style field cannot contradict itself, which is why the short-prompt people are happy, and a long one written by somebody who checks each word against the genre does not have to either.

The Suno team's own advice for the v5 model splits the difference in the useful direction: start with one or two genre tags, then expand the field one or two tags at a time.

That advice was written for v5, which Suno retired on 9 September 2026 along with every model before v6. On v6, users describe the opposite trouble with a short field: a two-word style comes back expanded into a paragraph of instruments and mix notes. Read the field before you generate, and cut what you did not ask for.

Put the genre first and make every other word agree with it

The check reads the field in order and expects the genre at the front. That ordering is a convention, not a documented behavior: guides such as HookGenius state that earlier descriptors weigh more, and Suno has not published anything that confirms it. What is not in dispute is the second half. Once the genre is chosen, every remaining word either supports it or fights it, and HookGenius keeps a useful list of pairs that fight, including country with 808s and rock with synth pads.

The rewrite that survived the findings above:

lo-fi bedroom pop, melancholic, acoustic guitar, female vocals, 90 bpm

Genre, one mood, the main instrument, the voice, a tempo. Five descriptors, no contradictions, and the check returns nothing to fix. The stadium chorus is gone because you gave it up on purpose rather than letting an average make the choice.

Tip

Change one thing per regeneration. If you rewrite the whole field between takes, you learn nothing about which word did what.

What the check cannot tell you about tags

It compares words against a conflict list. It has not heard the song, so a pair it calls a conflict can be exactly the record you want, and lo-fi with a polished mix is a real aesthetic in the right hands. It flags the disagreement and leaves the decision with you.

It also cannot make the Exclude field work. Users describe it as the least reliable part of the interface, one of them saying it ignores nearly everything and the exclusions most of all. An older complaint is sharper: asking for no autotune is effectively asking for autotune, which is what tends to happen when you name the thing you do not want.

And it says nothing about whether your five descriptors describe a song anyone wants to hear. That part is still writing.

Paste your lyrics, it takes ten seconds: heho.ai/tools/suno-lyrics-check