Directing vocals in a prompt
Vocals are where AI music most often goes off-script — a generator will happily add a loud, centred, generically-sung lead you never asked for. Treat the vocal like a session singer you're briefing, and you get far closer to what you hear in your head.
Name the voice
Register and character do the heavy lifting: "male baritone, conversational", "female, breathy, single-tracked", "androgynous, processed, distant". You don't need a name — you need the qualities that make the delivery specific.
Direct the delivery
How a line is sung matters more than who sings it. Half-spoken, belted, whispered, melodic-but-restrained. Say whether it's doubled (thicker, more produced) or single-tracked (intimate, exposed), and whether ad-libs sit wide in the stereo field.
Set the lyrical vibe in one phrase
You rarely need full lyrics — a direction is enough: "confessional, unguarded, no metaphors", "hard, present-tense, narrative", "about transcendence, not a specific person". This steers tone without locking you into words the model handles awkwardly.
Getting a clean instrumental
If you want no vocal, say "instrumental" explicitly, and consider naming a lead instrument to occupy the space a voice would ("melodica solo on the hook", "humming only, no words"). Left unsaid, many tools fill the gap with a vocal by default.
2,300+ deterministic prompts. Free to try.
Pick your decisions — mood, groove, instruments, energy — and the engine fills the rest in under a second. No card needed.
Try the engine →Frequently asked
Do I have to write the actual lyrics?
No. A one-line lyrical direction (subject + tone + point of view) usually steers the result better than full lyrics, which models can render awkwardly. Write full lyrics only when the exact words matter.
How do I stop the tool from adding a vocal I didn't want?
Say "instrumental" outright, and give the arrangement a lead element to carry the melody so there's no empty space for a vocal to fill.