If you picked a single vocalist but the song came back as a duet, nothing is broken. The model just needs a stronger signal, and there's a reliable way to give it one.

The vocalist gender dropdown isn't a hard lock. When the model builds a song, it weighs your genres, vibes, and lyrics fields more heavily than the dropdown on its own. So if anything in those fields reads as gender-ambiguous, the model can add a second voice and hand you a duet instead of the single voice you selected.

Frustrating? A little. Fixable? Almost always.

The trick is to repeat your intended vocal in more than one place, so the model has no room to drift. Reinforce it across four spots.

Leave Female (or Male) selected in the vocalist dropdown. This is your baseline, just not your only signal.

Near the front of your genres or vibes field, add a few matching vocal descriptors. The more specific, the better.

For a female voice, try:

For a male voice, try:

At the very top of your Custom Lyrics field, add this as the first line:

or the reverse:

Then reinforce it at the start of each section:

Run 4 to 6 versions of the song. This gives you the best odds of landing a clean single-voice take.

Even with all four layers in place, the model occasionally still slips in a second voice. Stacking reduces this dramatically, but it doesn't remove it completely. If a duet shows up anyway, just regenerate. It usually falls into line on the next pass.

Sometimes one pass is enough. Sometimes it takes two. That's normal.