Copy the four fields to Suno

Suno reads four separate inputs, and the single most common prompting mistake is collapsing them into one. A style description with lyrics pasted into it, or a lyric box carrying production notes, gives the model two jobs in one instruction and it will do neither cleanly.

Each generated card has exactly one destination.

Style

Goes in Suno's style description box.

This is the genre, tempo, instrumentation, vocal direction, production character and mood — compiled into one comma-separated line in a deliberate order, because the order is part of the instruction. Keep it there. Do not add lyrics, section tags or anything phrased as a negative; each of those has its own field below.

Lyrics

Goes in the lyric editor, with the section tags left intact.

The tags are not decoration. [Intro: stripped back], [Chorus: powerful vocals] and [End] are how you tell the model where the arrangement should change and where the song should stop. Deleting them because they look untidy is the most common cause of a track that never ends or that arrives at full volume from the first bar.

Exclude

Paste it only when it is not empty.

Exclude is for things you do not want to hear — auto-tune, heavy reverb, vocals on an instrumental. It is a genuinely useful field and an easy one to overload: every exclusion is another constraint competing with the positive instructions in Style, and a long exclusion list often does more damage than the thing it was meant to prevent. Exclude what actually ruined the last take, not everything you can imagine going wrong.

Settings

This one is a checklist, not a paste.

These are the values you set by hand in Suno's own web controls — the ones that live in the interface rather than in a text box. Work down the list and set each one. If an item does not appear in your Suno interface, that is expected: the models do not all expose the same controls, and the checklist reflects the model you chose to target.

Why keeping them separate matters

Two reasons, and the second is the one people underestimate.

The first is that the instruction is clearer. A model given a clean style line, tagged lyrics and an explicit exclusion list has much less to disambiguate than one handed a paragraph containing all three.

The second is that separation is what makes the next version diagnosable. When the four fields are distinct, a disappointing take points at one of them — the vocal is wrong, so the vocal direction in Style is wrong. When everything was pasted into one box, a bad result tells you nothing about which decision caused it, and you are back to rewriting the whole prompt and hoping. That is exactly the loop that one-variable iteration exists to break.