1. Introduction
← Guide contents · Next: Philosophy →
There is a point where adding another genre name stops helping me.
A prompt such as Liquid Drum and Bass, Chiptune, Spectral describes the neighbourhood, but it leaves most of the musical decisions unresolved. It does not tell me whether the chip timbres are the main identity or a surface texture, whether the breaks should roll or stop for halftime sections, or which part must survive when those ideas compete.
Semantic prompting is how I make those decisions visible. I separate the idea into roles—bass, rhythm, harmony, texture, motion, priority, structure—and give each one something useful to do. This does not give me perfect control. It gives me a clearer reason for writing each line and a better way to understand what needs changing after I listen.
A guide to thinking about generative music
Section titled “A guide to thinking about generative music”Despite its title, this is not really a guide to filling a prompt box. Prompts are one part of the process. I am more interested in how a musical problem can be split into manageable parts, how source material can carry an idea forward, and how one generation tells me what the next stage needs to do.
Some of the way I think about this comes from software engineering:
- Separation of concerns — let different prompt fields, source tracks or workflow stages solve different musical problems.
- Iterative refinement — revise in response to a result instead of trying to specify the finished song perfectly in advance.
- Provenance — keep the source, intermediate generations, settings and creator relationships visible.
- Documentation — record what each stage was meant to do, what happened and why the next decision followed.
Music is not software, and Suno is not a deterministic compiler. I am not expecting the same prompt to return the same track. These principles simply give the uncertainty some structure: I can tell which idea I was testing, what survived, and why I made the next change.
This guide assumes you already know your way around Suno. I am concentrating on the part I find more interesting: deciding what to ask for, hearing what the generation actually did, and working out why it needs another pass.
What this guide is not
Section titled “What this guide is not”This guide is not a reverse-engineering of Suno.
It is not a list of magic words.
It does not guarantee specific outputs.
It is a record of creative practices and observations that have been useful to me: ways of structuring musical intent, testing an idea and learning from the result.
Suno remains generative and variable. I am not trying to remove that uncertainty. I am trying to make better decisions inside it.
Workflows over prompts
Section titled “Workflows over prompts”I used to put too much pressure on the prompt to solve the whole song at once. That becomes especially obvious with genre fusion. One block of text may be asked to preserve a melody, invent a new rhythmic identity, carry two production languages and make the result feel cohesive.
Many of my strongest results have come from giving those problems to different stages. In Flowstate In Yellow, one track carried the vocal melody and atmosphere while another carried motion and tension. Mashup assembled them, then Cover made the result feel like one song. No single stage had to invent everything.
That is what I mean by workflows over prompts. The words still matter, but so do the order I use them in, the audio they act on, and the job I expect each stage to finish before moving on.