Music basics for prompting

Why you need this

A music model listens for musical words: a tempo, a key, a genre, a list of instruments, section names like verse and chorus. Some models even take tempo and key as separate settings. You do not need to read music to use them, but you do need to know what each word means and how tempo and bars turn into seconds, because a game loop has to be a whole number of bars long.

The idea

Beat and tempo. The beat is the steady pulse you tap your foot to. Tempo is how fast the beat goes, in BPM (beats per minute). One beat lasts

$$t_{\text{beat}} = \frac{60}{\text{BPM}} \text{ seconds.}$$

At 120 BPM a beat is $60 / 120 = 0.5$ s. At 90 BPM it is about 0.667 s. Rough feel: 60 to 80 BPM is slow and calm, 90 to 120 is moderate, 130 and up is fast and urgent.

Bars and time signature. Beats are grouped into bars (also called measures). The time signature says how many beats are in a bar and which note counts as one beat.

Time signature	Count	Feel	Typical use
4/4	1 2 3 4	Steady, square	Most pop, rock, electronic, and game music
3/4	1 2 3	Lilting, waltz	Waltzes, gentle folk, music-box themes
6/8	1 2 3 4 5 6	Rolling, two big beats each split in three	Jigs, sea shanties, many ballads

In 6/8 the listener usually feels two beats per bar, each divided into three quick pulses, so a "tempo" for 6/8 can mean either the quick pulse or the big beat. Say which when it matters. The length of a passage in seconds is

$$t = \text{bars} \times \text{beats per bar} \times \frac{60}{\text{BPM}}.$$

Musical sections are measured in bars on this site; everything else is in seconds.

Notes and octaves. Pitch is how high a note sounds, set by its frequency in hertz (vibrations per second). Western music names twelve notes per octave: C, C#, D, D#, E, F, F#, G, G#, A, A#, B, where # (sharp) means one step up. Some of those notes have a second name with b (flat) meaning one step down, so C# is also Db. A number gives the octave: A4 is the A at 440 Hz, A5 is an octave higher at 880 Hz, and doubling the frequency always raises the pitch one octave. Each of the twelve steps multiplies the frequency by $2^{1/12} \approx 1.0595$, so C4 (middle C), nine steps below A4, is $440 \times 2^{-9/12} \approx 261.6$ Hz.

Keys and mood. A key is the set of seven notes a piece mostly uses, named by its home note and whether it is major or minor. C major uses the white keys of a piano: C D E F G A B. A minor uses the same seven notes but treats A as home. Major keys tend to sound bright, settled, or triumphant; minor keys tend to sound sad, tense, or mysterious. This is a tendency, not a law: tempo and instruments change mood just as much.

Chords and progressions. A chord is several notes played together. The basic chord stacks every other note of the key: C major is C, E, G; A minor is A, C, E. A progression is a sequence of chords, written with Roman numerals for their position in the key, upper case for major chords and lower case for minor ones.

Progression	In C major	Common feel
I, V, vi, IV	C, G, A minor, F	Uplifting pop, anthems
vi, IV, I, V	A minor, F, C, G	The same chords starting sad, then resolving
I, IV, V, I	C, F, G, C	Simple, folk, a strong ending
i, VI, III, VII (in A minor)	A minor, F, C, G	Dark, heroic, very common in game soundtracks

You do not need to write these into a prompt.

Song sections. Songs are built from repeated blocks:

Section	Role	Typical length
Intro	Sets the mood before the voice enters	4 to 8 bars
Verse	Tells the story; lyrics change each time	8 to 16 bars
Chorus	The hook; same words each time; usually the loudest part	8 bars
Bridge	Something different near the end, to refresh the last chorus	4 to 8 bars
Outro	Winds down or ends	4 to 8 bars

The lengths are conventions, not rules. Instrumental game music often skips verses and choruses and uses an A section and a B section that alternate.

Instrumentation and genre. Instrumentation is the list of instruments: piano, strings, synth pads, acoustic guitar, drum kit, harp. Genre is a style label that implies a whole bundle of tempo, instruments, and production: "lo-fi hip hop" implies a slow, dusty drum loop and soft keys; "orchestral" implies strings, brass, and timpani. Models learned genres from labeled examples, so a genre word is often the most powerful single word in a prompt.

Worked example

How long is 8 bars of 4/4? There are $8 \times 4 = 32$ beats.

Tempo	One beat	8 bars
120 BPM	$60 / 120 = 0.5$ s	$32 \times 0.5 = 16$ s
90 BPM	$60 / 90 \approx 0.667$ s	$32 \times 0.667 \approx 21.33$ s

At 120 BPM the phrase lands on a round 16 s. At 90 BPM it is 21.33 s, which still loops perfectly as long as you cut it in samples rather than rounding to a tidy number of seconds.

Choosing the Lumen Clash lobby loop length. The lobby is where players wait between matches, so the loop should be long enough not to grate and short enough to keep the file small. Suppose you want 40 to 50 s of relaxed, moderate music in 4/4. Try a few tempos and bar counts:

Tempo	Bars	Beats	Length
90 BPM	16	64	42.67 s
96 BPM	16	64	40 s
120 BPM	24	96	48 s

120 BPM with 24 bars gives a round 48 s: three 8-bar phrases of 16 s each. That is the choice the rest of this cluster builds on. A bar at 120 BPM is $4 \times 0.5 = 2$ s, which is handy later when you need to wait for the next bar line before switching music. The in-match loop, which should feel more urgent, uses the same 24 bars at 150 BPM: $96 \times 0.4 = 38.4$ s.

In a game's audio pipeline

These words belong to stage 2, Write the brief: tempo, key, genre, and instruments go into the prompt or into a model's settings. Bars and tempo come back in stage 4, Shape it for the game, where a loop must be cut on a bar line and a win or lose cue should wait for the next bar so it lands in time with the music.

Common mistakes

• Mixing tempo units. You ask for 120 BPM in 6/8 and get a frantic track, because the model took 120 as the quick pulse. Use 4/4 unless you have a reason not to.
• Cutting a loop on whole seconds. You trim a 90 BPM loop at 21 s and hear a stumble every repeat. 8 bars at 90 BPM is 21.33 s; cut at the bar length.
• Expecting minor to mean sad. A fast minor-key track with driving drums sounds exciting, not sad. Mood comes from tempo and instruments as much as key.
• Over-specifying chords. You write a chord progression into the prompt and the model ignores it or garbles the result. Most models respond to genre and mood words far more reliably than to chord names.
• Forgetting the intro. A song model often spends 4 to 8 bars on an intro, so a 16 s request for an 8-bar loop may come back mostly intro.

Cost

The vocabulary costs nothing but reading time. The choices it describes do cost something: loop length sets file size and listening time. The 48 s lobby loop, stored uncompressed at 48 kHz (48,000 samples per second), 16-bit, stereo, is $48{,}000 \times 2 \times 2 \times 48 = 9{,}216{,}000$ bytes, or about 9.2 MB with 1 MB = 1,000,000 bytes. Halving the bar count halves that, and halves the time it takes you to listen to a take, at the price of a loop players notice repeating sooner.

Going further

• Prompting for music, which turns this vocabulary into prompts.
• Seamless music loops, which cuts loops on the bar lines computed here.
• Any beginner music theory text, for the circle of fifths and why I, IV, V sound so stable together.