How to Write AI Song Prompts: Framework + Examples

A music prompt has to describe both taste and job. Taste tells the model the sound world: pop, folk, electronic, cinematic, acoustic, upbeat, intimate, dark, warm, polished, raw. Job tells the model why the song exists: a birthday reveal, a YouTube opener, a product launch hook, a podcast intro, a chorus-first demo, or a private draft.

When a prompt only names taste, the model has too many open choices. "Create an upbeat pop song" can become a dance track, a teen-pop chorus, a commercial jingle, or a full vocal song with busy lyrics. When the prompt names the job, the model can shape duration, arrangement, vocal density, and ending behavior around the intended deliverable.

The formula in one line: use case + subject + genre + mood + vocal + ending. Write those six as a single short paragraph, not a form.

A reliable GoAISong prompt can follow this order: use case, subject, genre, mood, vocal and arrangement, ending. That order is useful because it starts from the product goal before moving into style. You can write it as one paragraph, not a form. The point is to make each sentence do a job.

For example: "Create a 45 second warm indie pop song for a small creator launch video. The lyrics should be about starting something new after months of preparation. Use bright guitars, light drums, a female vocal, and an optimistic chorus hook. Keep the verses short, avoid real-artist references, leave a clean ending for a title card." This is specific without becoming brittle.

Description-first prompting is best when you know the idea but not the exact words. You describe the song, and GoAISong drafts lyrics, melody, vocals, and production together. This is useful for quick concept testing, personal songs, creator hooks, and first-pass demos.

Lyrics-first prompting is better when the words matter more than the model invention. If you already have a poem, verse, brand line, wedding speech, or chorus, use the custom lyrics workflow. In that case, the prompt should focus less on story and more on musical treatment: tempo feel, genre, vocal tone, and how the lyrics should be paced.

Genre and mood are not the same instruction. Genre controls the musical vocabulary: instruments, rhythm, production style, and common song shapes. Mood controls how that vocabulary feels. "Acoustic folk" can be nostalgic, hopeful, intimate, or tense. "Electronic" can be bright, dark, playful, minimal, or cinematic.

Most bad prompts either use one generic genre or overload the model with conflicting labels. A better pattern is one primary genre, one secondary texture, and one clear mood. "Warm acoustic folk with subtle cinematic strings, reflective but hopeful" is easier to execute than "folk pop country cinematic EDM inspirational sad happy."

Vocal direction should describe the role of the voice: intimate storyteller, confident pop lead, calm narrator, energetic rap delivery, soft background texture, or no lead vocal. That gives the model usable musical information without asking for imitation.

Do not name real artists, public figures, or celebrity voices as targets. Aside from policy and rights concerns, those references often make prompts less precise. "A clear female vocal with warm phrasing and a hopeful chorus" is more actionable than a comparison to a famous singer because it tells the model the performance job.

You do not need to write a formal score to control structure. Use plain edit language: quick intro, short verse, memorable chorus, bridge lift, clean ending, loopable middle, no long fade, room for voiceover, title-card stop. These instructions are especially useful when the song will sit inside a video, ad, podcast, or presentation.

If the output needs to be a full song, ask for a complete arc. If it needs to work as a cue, ask for a compact section with a clear entry and exit. A 30 to 60 second creator opener should not behave like a three-minute radio song; a gift song may need more lyrical development and a stronger chorus payoff.

A prompt can mention the intended use, such as "for a monetized creator video" or "for a client product launch draft," because that helps the song function as a deliverable. But the prompt itself does not grant rights. Commercial use depends on the GoAISong license state for the generated output and on the rules of the destination platform or client contract.

The practical habit is to separate creative prompting from rights packaging. First generate drafts. Then, when you choose the song that will appear in a commercial-use context, make sure the account has Creator or Studio commercial-use coverage and keep the certificate or receipt with the project record.

Use these as starting points. Each one follows the six-part formula, names a use case, and stays free of real-artist references. Swap the names, details, and genre to fit your project, then paste it into the generator.

The easiest way to improve a prompt is to add the missing job. If your first draft is "make a happy song," ask what the song is for. A wedding first dance, a birthday surprise, a gaming channel intro, and a product demo can all be happy, but each needs a different arrangement and ending.

The second improvement is to remove imitation language and replace it with musical traits. Instead of asking for a real artist, describe tempo feel, vocal tone, instrumentation, lyric angle, and mix density. That gives GoAISong a cleaner target and keeps the generated song attached to your own brief.