Estimated time: 75 minutes. Competencies: prompt engineering, creativity and innovation.
Video: Lesson 1.1: Composing an Original Song with GenAI. Three-minute introduction: the five-part prompt frame, structure tags, and iterating one variable at a time. (Instructor: build with make_videos.py or record and embed your own.)
Introduction: Music Lesson One. Recorded by the instructor for the 2025 pilot of this program.
Learning objectives
By the end of this lesson you will be able to:
Write a structured song prompt that specifies genre, tempo, mood, instrumentation, vocal style, and form.
Generate at least three versions of a song from your own lyrics and select one using musical criteria.
Explain, in a 75-word rationale, which prompt choices produced the strongest result and why.
Warm-up: listen like a producer (5 minutes)
Play any recorded song you know well. While it plays, write down six things a producer had to decide: tempo, key or mode, instrumentation, vocal style, form (verse, chorus, bridge), and mood. These six decisions are exactly what a text-to-song model needs from you. If you leave them out, the model decides for you.
Key ideas
A prompt is a brief, not a wish
Text-to-song tools such as Suno take two inputs: a style prompt (a short description of how the song should sound) and lyrics (the words, with structure tags such as [Verse] and [Chorus]). Vague prompts ("a nice pop song") produce generic output. Specific prompts ("upbeat indie pop, 118 BPM, bright acoustic guitar and handclaps, warm female alto vocal, sparse first verse building to a full chorus") give the model constraints to work within, and constraints are where style lives.
The five-part prompt frame
Use this frame for every music prompt in the course:
How does it change over time, and how should it feel?
"starts hushed, opens up at the second chorus, bittersweet"
Structure tags do the sequencing
Lyrics with tags such as [Intro], [Verse 1], [Pre-Chorus], [Chorus], [Bridge], and [Outro] tell the model where sections begin. Without them, choruses land in strange places. Tags are your first taste of computational thinking: you are sequencing the form so the model can follow it.
Iterate on purpose
Generation is cheap; judgment is not. Change one variable per generation so you can hear what it did. A useful order: get the form right first, then the tempo and feel, then the instrumentation, and finally the vocal.
Activity: compose your song (45 minutes)
Write lyrics for a short song: two verses and a chorus, optionally a bridge. Keep it under 16 lines. Choose a topic connected to teaching, your community, or a piece you already love. Write in your own words; do not ask a text model to write the lyrics for this lesson.
Tag the structure with [Verse 1], [Chorus], [Verse 2], [Chorus], [Bridge], [Chorus].
Write a style prompt using the five-part frame. Keep it under 200 characters if the tool limits length.
Generate version 1 in Suno with custom mode on, pasting your lyrics and style prompt.
Listen critically. Fill in the evaluation grid below.
Change one variable and generate version 2. Repeat for version 3. Record what you changed each time in your Music Module Notes.
Select the strongest version. Download it as MP3 and name it Lesson-1-1_LastName_Title.mp3.
Evaluation grid
Criterion
Version 1
Version 2
Version 3
Form follows my tags
Tempo and feel match the brief
Instrumentation matches the brief
Vocal phrasing is natural and intelligible
Lyrics are sung as written
Overall: would I play this for a class?
Prompt bank
Adapt any of these as starting points:
"Gospel-influenced pop, 96 BPM, piano and choir, powerful female lead, builds to a big final chorus, hopeful"
"Lo-fi hip hop, 82 BPM, dusty drums and vinyl crackle, soft male spoken-sung vocal, mellow and reflective"