Blog

Speech reads your script at its own pace, and the first complaints are about words read wrong and pauses in the wrong place.

The closest-template match is sixty percent section order, thirty-five percent genre and twenty percent mood, which is worth knowing before you read it as a verdict.

A model asked to continue your first verse has one thing to work from, so the most probable second verse is a paraphrase of the first.

Digits are invisible to a syllable count and ambiguous to a singer, and writing them out fixes both problems at once.

Every problem you find by listening cost you a generation, and most of them were visible on the page.

Changing one thing and listening once tells you almost nothing, because the two takes differ for reasons that have nothing to do with your change.

A misspelling is a word Suno has never seen, so it guesses, and the guess is stable across regenerations.

Every style descriptor that works pairs a feeling word with a physical noun, and an artist name is the feeling word standing on its own.

Without tags there is nothing telling Suno where the song changes, so it does not change.

Exclusions are guidance rather than bans, and naming what you do not want is a weaker move than describing what you do.

A line is measured in bars, not seconds, so the syllable budget stays put while the tempo decides how fast you have to deliver it.

Two voices in one generation is contested, and the workflow that people report succeeding with builds each voice separately.