-
01
How music models read a prompt
≈ 3 minA music model listens for genre, mood, tempo and instrumentation - in that order. One precise line beats a poetic paragraph: uplifting lo-fi, warm piano, 90 BPM, no vocals.
PracticeGenerate one track from a single line that names genre, mood, tempo and one instrument.
Generate your first track → -
02
The vocabulary: genre, tempo, instruments
≈ 4 minBPM sets energy: 60-80 calm, 90-110 neutral, 120+ drive. Naming two instruments gives the model an anchor. Naming five gives it noise.
PracticeGenerate the same mood twice: once at 70 BPM, once at 125 BPM - hear what tempo alone changes.
Test two tempos → -
03
Structure for 60-second beds
≈ 4 minBackground music must stay in the background. Ask for a steady bed with a subtle lift mid-way and a clean ending - not a chorus. No vocals is the most useful phrase in this course.
PracticeGenerate a 60-second bed with a mid-way lift and a clean ending for a product video.
Build a 60s bed → -
04
Voiceover on top: mixing speech and music
≈ 5 minA voiceover needs space: calmer beds, softer highs, and the voice generated separately so you control both volumes. One minute of narration is about 1,000 characters.
PracticeWrite a 3-sentence script, generate the voiceover, and play it over your lesson-3 bed.
Add a voiceover → -
05
Rights: what the commercial license covers
≈ 3 minWith a paid credit pack you can use what you generate commercially - ads, client work, monetized videos - within each model's license, shown on its model page. Rates are flat and published, so you know what a soundtrack costs before you make it.
PracticeCheck what a credit pack costs and where the 25-credit track rate sits.
See credit packs & the license →
Prompt, structure, mix, rights - the full soundtrack loop. Score your next video from a single line.
Generate a track now →Finished a course? Take a 20-question test and get a free, verifiable AI certificate.
Take the free test →How to work through this
Each lesson ends in a task you can run in about two minutes. The reading is the shorter half. The part that teaches you something is the run, because the result argues back.
Do the tasks in order. Later lessons assume the settings from earlier ones. Skipping ahead usually means an expensive model on a prompt that was not ready.
Keep the link from every run you liked. It carries the model and the settings, so a good result is repeatable without remembering how you got there.
What the course costs to follow
Reading is free and needs no account. Only the tasks cost credits, and a new account starts with 100 of them without a card.
Free credits cover the tasks that run on fast image models. Tasks on video, music, voice and premium image models need a credit pack. If a lesson suggests an expensive model, it also says what the cheaper draft version is.
Each task shows its credit cost before you run it. A practice run never costs more than you expect.
Questions about
Do I need an account to read this?
No. The lessons are open. An account is needed only to run the tasks, and it starts with 100 free credits.
How long does the course take?
About 19 minutes, tasks included. Reading alone is quicker, but reading without running is how people forget it a week later.
Do I need any experience?
No. The first lesson assumes you have never written a prompt. The later ones assume you did the earlier tasks.
Can I use what I make here?
Yes, on paid packs, under the UGESI terms and the license of the model that produced the result.
