How to Structure a Mureka Style Prompt

Mureka's own documentation contains the best style prompt example anyone has written about it, and almost nobody quotes it. Here it is in full: r&b, slow, passionate, male vocal.

Four fragments, comma separated, no sentences. A genre, a tempo, a mood, and a vocal direction. That is the whole template, and everything below is just an expansion of it.

What the field actually is

In Mureka's song generation API, the style prompt is a field called prompt, described as "Control music generation by inputting a prompt, maximum 1024 characters". It sits next to a required lyrics field that is documented at up to 5000 characters. Mureka's Easy mode has its own prompt field documented at 2000 characters.

Two things follow from that. First, the style prompt is a description of the sound, not of the song's content, because the content already has its own field. Second, 1024 characters is a lot of room compared to how much you actually need. Mureka's example uses about thirty of them. Length is not the constraint. Precision is.

I should say plainly where my knowledge stops. The limits and the field behaviour above are from Mureka's published docs. I have not run generation tests on Mureka, so I am not going to describe how any given prompt renders there. The structure below is what Mureka's docs imply plus what I learned writing style prompts on Suno, where I have done the reps.

The five slots

I write every style prompt as the same five slots, in the same order, comma separated.

1. Genre, one or two words. Not a stack of five. "Indie folk" is a genre. "Indie folk rock alternative acoustic americana" is a shrug.
2. Tempo and energy. A word and a number if you have one. "Slow", "mid tempo around 92 bpm", "driving". The number is not magic but it removes an ambiguity that adjectives leave open.
3. Instrumentation, two to four named instruments. Actual instruments. "Brushed drums, upright bass, clean electric guitar" tells the model what to build. "Rich instrumentation" tells it nothing.
4. Vocal direction. Range, delivery, and grit. "Restrained male vocal", "belted female vocal with rasp". On Mureka you can also set vocal gender in its own dedicated field, separate from the prompt, so the prompt text can spend its words on delivery instead.
5. Production and mix. How it should be recorded, not how good it should be. "Dry close mix", "tape saturation", "room reverb on drums only".

Then I stop. No sixth slot for adjectives about quality.

The rule that fixed my prompts

Nouns over adjectives. Almost every bad style prompt I have written was a list of compliments the model cannot act on.

"Emotional" is not a sound. "Cinematic" is not a sound. "Professional", "high quality", "masterpiece", "beautiful production" are all instructions to try harder, and no generator has a try-harder setting. Meanwhile "upright bass", "88 bpm", "close mic", "no reverb" are all things a mix either has or does not have.

The test I use: could a session musician act on this word without asking a follow-up question. "Brushed drums" passes. "Powerful" does not.

Before and after

Before, style prompt:

epic emotional ballad, incredible vocals, amazing production, cinematic, powerful, moving, high quality, professional sound, best song ever

Nine fragments, and only one of them ("ballad") describes anything. The rest are grades. If you handed that to a band they would all play something different and all of them would be right.

Before, lyrics:

[Verse]
I have been carrying this letter in my coat since the second week of February and I still have not decided whether the right thing to do is post it or burn it
Winter
The gate is still open

[Chorus]
Say it
Say it
I never said any of the things I meant to say to you when it would have actually changed how any of this turned out for either of us

The lyric block has tags, which is a start, but the lines inside are all over the place. Line one of the verse is over forty syllables next to a one-word line. The chorus does the same thing in reverse, two one-word lines then a long run.

After, style prompt:

folk ballad, slow, around 68 bpm, fingerpicked acoustic guitar, upright bass, low strings pad,
restrained male vocal with breath audible, dry close mix, minimal reverb

After, lyrics:

[Verse]
I've carried this letter since February
Folded in the pocket of my coat
Some days I mean to post it
Some days I mean to burn it

[Chorus]
Say it, say it, say it now
I never said a word back then
Say it, say it, say it now
Back when saying it would've meant something

The prompt now names a genre, a tempo with a number, three instruments, a vocal delivery, and two mix decisions. It is longer in characters and shorter in ambition, which is the trade I want.

The lyric block is doing its half of the job too. Verse lines sit between seven and eleven syllables each. The chorus has a repeated line, which is the simplest way to make a hook behave like one, and the long trailing line is now the payoff at the end of the section rather than a wall in the middle of it.

A few things worth knowing before you spend credits

Mureka's docs list a handful of controls that live outside the prompt text, and using them keeps the prompt clean.

• Vocal gender is a separate field with two values, so you do not have to spend prompt words on it.
• The model field accepts auto, which the docs describe as selecting the latest version of the regular model. That is what I would leave it on.
• A reference audio upload is documented in Custom mode, and Mureka says it extracts style tags from the file, giving an example of "Indie Pop, Acoustic Guitar, 110 BPM". If you have a track that already sounds right, that is a shortcut past describing it.
• Instrumentals have their own endpoint that takes a prompt and no lyrics, so a purely instrumental idea does not need a lyric sheet at all.

Why I built Mugical

Writing the lyric sheet and then writing a matching style prompt, by hand, every time, is the part of this that made me build something. You give Mugical a song idea plus genre, mood, instrumentation and era, and it produces both outputs from those same inputs, so the tempo and instrumentation language in the style prompt is describing the same song the lyric sheet is shaped for, rather than two songs that happen to be open in the same tab.

It does not generate audio. That step still belongs to whichever platform you paste into. What it hands you is a lyric sheet and a style prompt built to agree with each other.

You can try it on a song idea and look at the pair side by side. If you want the wider picture of what Mureka takes as input, I covered that in a practical Mureka prompt guide. The other half of the job, the words themselves, is in how to write lyrics for Mureka.