How to turn a song into a prompt

A music prompt is just a precise description of a piece of music. The fastest way to write a good one is to start from a song that already sounds the way you want, then describe what is actually in it. That is what reversing a track does for you automatically: it turns audio back into the words that could have produced it.

This guide covers both routes. The manual way teaches you what matters and trains your ear, so you understand the output rather than trusting it blindly. The automatic way uses the reverse tool to measure the parts you cannot judge exactly by ear, tempo and key especially, and to draft the descriptive layer for you. Most people end up using a blend: measure with the tool, then edit with their own judgement.

Why start from a song at all

Describing music from scratch is hard because you are translating a feeling into vocabulary you may not have. Starting from a reference track flips the problem around. Instead of inventing a description, you are reading one off something real. The track already contains every answer you need: a fixed tempo, a definite key, a specific set of instruments and a recognisable production style. Your job is just to name them. This is the same trick session musicians use when a client says make it sound like this, and it works because matching is far easier than imagining.

It also sidesteps the most common failure in prompting, which is describing a mood instead of a sound. A generator cannot act on the word chill, but it can act on 72 BPM, a Rhodes piano, brushed drums and tape saturation. A reference track forces you down to that concrete layer because those things are simply present in the audio.

The five things a prompt needs

Most weak prompts fail because they describe a vibe instead of a sound. Cover these five dimensions and you have a usable prompt. They are listed in roughly the order of how much they shape the result, so if you only get the first three right you are already most of the way there.

• Genre and subgenre, for example lo-fi hip-hop rather than just hip-hop, or UK drill rather than just rap. The subgenre carries far more sonic information than the parent genre.
• Tempo in BPM, or a clear feel like slow, mid-tempo or driving. Tempo controls energy more reliably than any adjective.
• Key and mode if emotion matters, for example A minor for a darker feel, F major for something warm and open.
• Instrumentation, the actual sounds: electric piano, sub bass, brushed drums, analog synth arpeggios. Name the two or three that define the track, not everything in the mix.
• Mood and texture: nostalgic, warm, vinyl crackle, tape saturation, wide stereo. This is the finishing colour, applied after the structural choices.

Reverse a track step by step

1. 1

   Upload or record the track

   Open the reverse tool and drop in an audio file, or record a clip from a speaker or your library. Analysis runs in your browser, so a short representative section, twenty to thirty seconds of the main groove, is plenty. Avoid the intro or a sparse breakdown, since tempo and key detection work best where the full arrangement is playing.

2. 2

   Read the detected values

   The tool measures tempo, key, mode, duration and energy directly from the audio. These are facts about the file, not guesses, so trust them as your foundation. If a value looks surprising, see the note below on octave and tempo ambiguity, which explains the handful of cases where a measurement is technically correct but not what you expected.

3. 3

   Check the interpreted layer

   Genre, mood, descriptors and instruments are inferred, the model's best read of the sound. Treat these as a strong starting point you can correct rather than gospel. The model is reading texture and arrangement, so it is usually close on family and feel, and occasionally off on a specific instrument that sits low in the mix.

4. 4

   Edit anything that is off

   Swap a genre, add an instrument, halve or double the BPM, or flip to the relative major or minor key. The prompt text recomposes as you edit, so you always see exactly what you are about to copy or generate. This is where your own ear earns its keep.

5. 5

   Copy or generate

   Copy the finished prompt to use anywhere, or generate a new track from it here on a licensed model. You can also send it straight to the lyrics or enhancer tools to build the rest of the song around it.

A worked example

Say you reverse a relaxed instrumental and the tool detects 74 BPM in F major with low energy. It interprets the genre as lo-fi hip-hop with a nostalgic mood and electric piano. Composed into a prompt, that becomes:

Lo-fi hip-hop, ~74 BPM, F major, jazzy electric piano chords, mellow boom-bap drums, warm sub bass, vinyl crackle, nostalgic and relaxed, instrumental.

From here you change one thing at a time. Drop to 68 BPM for a sleepier feel, add rain and distant city ambience for a rainy-day version, or swap the drums for brushed jazz drums to make it more organic. Small, concrete edits move the output reliably. Changing several things at once makes it impossible to tell which edit did what, so resist the urge to rewrite the whole line.

When a measurement looks wrong

Tempo and key are measured accurately for the large majority of music, but two honest ambiguities exist and it helps to recognise them. The first is octave error in tempo: a track at 140 BPM and the same track read as 70 BPM are mathematically both valid, because half-time and double-time share the same underlying pulse. If a number feels off by exactly double or half, that is what happened, and one tap fixes it. The second is the relative major and minor pairing: A minor and C major share the same notes, so a detector can land on the relative of the key you would name. Flip it and trust your ear. Beatless, rubato or heavily ambient pieces are the genuinely hard cases, and there the detected values are best treated as a suggestion.

Doing it by ear, without the tool

If you want to write a prompt from scratch, listen for the same five things in order. Tap your foot along to the beat and count the taps in fifteen seconds, then multiply by four to estimate the BPM. Hum the lowest note the music keeps returning to, the one that feels like home, to sense the key. Then name the loudest two or three instruments, the ones you would notice if they dropped out. Describe texture last, because it is the easiest layer to hear and the easiest to over-describe. The reverse tool simply removes the guesswork from tempo and key, which are the two values people get wrong most often by ear, and frees you to spend your attention on the parts only a human judges well.

Common mistakes when reversing

• Sampling a sparse intro or outro instead of the main section, which can throw off tempo and energy detection.
• Accepting the interpreted instruments without listening, when a quick check would catch a guitar labelled as a synth.
• Editing five things at once, so you cannot tell which change improved or broke the result.
• Keeping the descriptive line so long it contradicts itself, for example asking for both minimal and lush in the same prompt.
• Forgetting to keep instrumental in the prompt when the reference had no vocals, then being surprised by an added vocal.