THE STUDIO NOTEBOOK / 01
A clear idea makes
a better starting point.
Treat a prompt like a short brief for a musician: give it a direction, a feeling and a purpose. Then listen and adjust. These are starting points, not guaranteed recipes.
1. Choose the right input mode
Use Simple when you want the model to interpret a song idea, including its lyrics. Use Custom when you already have words: place those words in Lyrics and put the sound in Style. For a backing track, enable Instrumental and leave sung lyrics out.
2. Replace vague adjectives with musical choices
“Make an amazing happy song.”
“Uplifting indie pop about taking the last train home and starting again. Warm vocals, shimmering guitars, steady drums and a hopeful chorus.”
The second prompt gives a story and a small set of compatible sound cues. There is no need to name a famous artist or request an imitation.
3. Give a lyric room to be sung
A line such as “I remember every single little thing that happened on that rainy afternoon” can feel rushed. Try “I kept that rainy afternoon / Folded in a note from you”. Shorter lines and related images leave room for melody.
Mark sections with [Verse] and [Chorus], and repeat the actual chorus when you want it repeated. Preview the result for skipped words or unusual pronunciation. Structure labels remain guidance.
4. Tell background music what to leave out
For a podcast bed, focus on a sparse arrangement, soft drums and a steady groove. Select Instrumental. Avoid asking for a powerful lead vocal or a huge chorus at the same time.
5. Adjust the part that actually failed
| What you hear | What to change next |
|---|---|
| The lyrics feel rushed | Shorten lines and reduce syllables; keep the style the same for comparison. |
| The music feels cluttered | Remove competing genres and ask for two or three main instruments. |
| The energy is too high | Replace “anthemic, explosive” with “restrained, soft dynamics, relaxed groove”. |
| The instrumental competes with speech | Ask for a sparse arrangement and fewer foreground melodies; lower the music in your editor. |
| The ending is too short | Try the paid extension tool on a suitable result. It continues the song; it does not rewrite a precise section. |
6. Compare versions before spending more credits
Listen to each returned track in full. Note one change to try, keep your favorite and regenerate deliberately. Each ordinary generation task uses one paid credit; the cost is per task, not per returned track. Model changes and new generations can change the melody, voice and arrangement.
There is no guarantee of exact tempo, duration, identical voices or precise phrase replacement. Finish timing, levels and edits in your audio workstation where needed.