How to Write Suno Prompts: Six Elements and Four Mistakes
A practical framework for writing Suno style descriptions that deliver consistent results: genre, tempo, vocals, instruments, structure, and mix.
👁 124 reads
Suno accepts two fields: style description and song lyrics. The entire quality of the output is determined by the first one, and most advice online boils down to listing a genre and a mood. That’s not enough. Below is the framework we use, built from things we’ve verified in our own generations.
Six Elements of a Style Description
Order matters, because the model takes the beginning of the description more seriously than the end.
1. Genre and language. “Modern hip hop in Polish”, not “rap”. Specifying the language in the style description, not just in the lyrics, noticeably improves pronunciation.
2. Tempo. A specific number of beats per minute. “142 BPM” works, “fast” means nothing.
3. Vocals. Gender, timbre, delivery style. “Male rap vocal, clear diction, confident understated delivery” describes three different things at once and each one is controllable.
4. Instruments and production. List what should be audible: “deep 808 sub bass, crisp hi-hat rolls, minor piano loop”. It pays to be detailed here.
5. Structure. What happens in the verse versus the chorus. “Breathy intimate verses, huge anthemic chorus” tells the model where to be quiet and where to be loud. This is the element that most improves results and is most often missing.
6. Mix. “Glossy contemporary mix, radio ready master” or the opposite: “dry mix, no reverb on vocal, close-mic recording”. This determines whether the track sounds like it’s from the radio or from a room.
What Not to Do
Don’t name specific artists. Rights issues aside, it simply doesn’t work well: you get a caricature instead of a style. Describe the characteristics you want to capture, not the source.
Don’t write adjectives without substance. “Epic”, “emotional”, “amazing” carry no technical information. “Half-time atmospheric verses” does.
Don’t count on the genre name to handle genre-specific techniques. With K-pop, we had to explicitly list layered backing vocals, a built-up pre-chorus, and a sudden chorus opening separately, because the word “K-pop” alone didn’t trigger them.
Don’t forget about autotune. The model adds vocal tuning by default. If you’re making rap or anything raw, add “no autotune”.
Non-English Characters and Pronunciation
You write the lyrics themselves in your target language, with full diacritics. The style description is worth keeping in English, because production terms have established forms there. In the description, add “[language] lyrics with natural pronunciation”: this doesn’t guarantee perfect diction, but it clearly helps.
If you want to check pronunciation, weave difficult consonant clusters into the lyrics. We tested it in Polish using a verse with challenging sounds, because if that passes, almost anything will. You can hear the results in the article with three Polish tracks.
Tags in Song Lyrics
In the lyrics field, square brackets describing sections work: [Verse], [Chorus], [Bridge], [Outro]. This isn’t cosmetic, it’s a way to control structure: without them, the model decides on its own what’s a chorus.
Things like a countdown before the chorus are better entered directly in the lyrics as a line, rather than requesting them in the style description. The model then treats them as material to be sung.
Which Model for What
| Model | When to Use |
|---|---|
| v6 | Default choice. But note: it rewrites the prompt in its own way, separately for each variant, so two results from one click can differ significantly. |
| v6-wild | For finding non-obvious solutions. In our tests it followed the description literally, which paradoxically makes it more predictable in that one aspect. |
| v6-mini | Available to everyone, even without payment. Perfectly sufficient for testing and learning to write descriptions. |
Three Details That Save Time
Tracks from v6 come out short, in our tests from 53 seconds to 1:19. If you need full length, you have to ask for it or extend the track.
The lyrics field doesn’t clear the way you think. It’s an editable area, not a regular input box, so when working in series, visually check what’s actually in it before you click.
Save descriptions that worked. Suno lets you save them in the tool, and a good style description is worth more than a single successful track, because you can reuse it. The basics of writing prompts, common to all tools, are covered in the article about what a prompt is.
Describing the sound is one thing, and the lyrics to be sung are another. We wrote about the latter separately: how to write song lyrics for AI.
More on this from QUANTSLANT
guides·AI Image Generator: How to Create Images from Text PromptsWe explain how to prepare an image from a description in Firefly or ChatGPT, with ready-to-use prompts and criteria for evaluating generated images.