Two songs can share a genre tag and still sound nothing alike, because genre is the least interesting thing you can tell a music model. What moves the result is the arc: where it starts, what layers in, where it peaks and how it ends. Listing "melancholic indie" gives the model a shelf to stand on; describing a whisper that builds into a full-throated chorus and then strips back to one piano gives it a performance.
This builder asks for a genre family, a mood, a vocal character, a production era, a mix texture and a dynamic arc, then writes the style text, the exclude list and a lyric skeleton with bracketed performance tags. It also holds you to the discipline that keeps output musical: the vocal character matters more than any single tag, and past roughly eight tags per section the directions start cancelling each other out.
Build a prompt
Nothing here talks to a server: the wording is assembled in your browser from the choices below, so it costs nothing and works with an ad blocker on.
Style field, exclude list and a lyric skeleton for Suno-style custom mode.
Only include a tempo if you actually want to hold the song to it. Ballads sit near 60–80, most pop is 90–120, dance is 120–128.
Featured instruments
Pick up to four. Each one you add is another instruction competing for the model's attention.
Keep out of the mix
This list becomes the exclude field. A short exclude list works; a long one starts removing things you wanted.
Style field
cinematic orchestral score, sweeping strings and low brass, tympani swells, melancholic mood, modern clean production, lead vocal, a weathered torch singer with a smoky alto and a slight rasp, warm analog tape saturation, 96 BPM. Opens restrained and repetitive, adds one new element every section, and peaks only in the final third before settling into a long, quiet outro.
Paste this whole thing into the style or genre box. It is written to stay inside a 1,000-character style limit, and it describes the journey, not just the genre.
Exclude styles
—
Optional, and short. Every word here is something you asked not to hear, so it is worth trimming if the result already obeys.
Lyric skeleton (custom mode lyrics box)
[Intro] [Female Vocals] (line one — the image or detail) (line two — what it changes) (line three — the turn) (line four — the hook line) [Verse 1] [Whispered] (line one — the image or detail) (line two — what it changes) (line three — the turn) (line four — the hook line) [Chorus] [Building Energy] (line one — the image or detail) (line two — what it changes) (line three — the turn) (line four — the hook line) [Verse 2] [Powerful] (line one — the image or detail) (line two — what it changes) (line three — the turn) (line four — the hook line) [Chorus] [Harmonies] (line one — the image or detail) (line two — what it changes) (line three — the turn) (line four — the hook line) [Bridge] [Female Vocals] (line one — the image or detail) (line two — what it changes) (line three — the turn) (line four — the hook line) [Chorus] [Whispered] (line one — the image or detail) (line two — what it changes) (line three — the turn) (line four — the hook line) [Outro] [Building Energy] (line one — the image or detail) (line two — what it changes) (line three — the turn) (line four — the hook line) [Silence]
The bracket tags are performance directions. Replace the placeholder lines with real lines, keep the tags, and delete a tag if you dislike what it does. Five to eight tags per section is the useful ceiling — past that the directions start cancelling each other out.
No account, no API key, no generation. This page only writes the text — you paste it into whichever tool you already pay for. Nothing you type here is uploaded, and this site is not affiliated with any generation service.
How to use
- Choose a genre family and a mood. If the two disagree, the mood usually wins in the output.
- Describe the vocal, including who is singing it — "a weathered torch singer with a smoky alto" steers more than "female vocals".
- Pick a dynamic arc. This is the control that changes the result most, so it is worth comparing two arcs on the same settings.
- Add up to four featured instruments, and only the ones you would notice if they were missing.
- Copy the style text into the style field, the exclude list into the exclude field, and paste the lyric skeleton into the lyrics box in custom mode — keeping the bracket tags and replacing the placeholder lines.
Frequently asked questions
Do I need a subscription to a specific music service to use this?
No. This page writes text only. The style field, exclude list and bracket tags it produces are the standard inputs for custom-mode generation on the major AI music tools, and you paste them into whichever one you pay for.
Why should I not name a real artist or band?
Because most tools reject or quietly water down prompts that reference protected names, so you get a worse result and a rule you are breaking. The vocabulary here avoids them on purpose and describes the sound instead: era, instrumentation, vocal character, production texture.
What are the bracketed words in the lyric skeleton?
Performance directions. [Whispered], [Building Energy] and [Belted] tell the generator how to sing a section. Keep five to eight per section; contradictory pairs like [Calm] and [Aggressive] in the same section confuse the model rather than contrasting nicely.
How many generations should I expect to need?
Plan for a few attempts per keeper, and treat them as takes rather than failures. If two runs in a row miss, change the arc or the vocal character rather than adding more adjectives.
Should I include a BPM and a key?
Only if you are holding the song to them. A tempo range is a real constraint — a 70 BPM ballad and the same words at 120 BPM are different songs — so include it when you mean it.
Does the character limit matter?
Yes, on most custom-mode style fields, where longer descriptions are truncated. The generated text is written to stay inside a thousand characters including the arc sentence.