Short-song pipelines often collapse “write a prompt” into one box, even when the inputs are not the same type of signal. Lyrics are a verbal deliverable. Style is a set of musical constraints. A video still is a visual constraint that may not need a paragraph at all. This how-to keeps those lanes named. A workspace like WAV AI helps when you pick the control surface before you generate.
Map the deliverable to a control surface
| What you already have | Control surface | Do not start with |
|---|---|---|
| Locked lyrics plus a style you can name | Separate lyric and style fields | One sentence that restates the verses and the genre together |
| A scene-level description, no locked words | A single whole-song prompt | Empty lyric fields you feel obliged to invent |
| A video still, cover, illustration, or photo that already sets the mood | The image first, optional written note second | A long prompt that only describes the picture |
Andrea Agostinelli et al., MusicLM: Generating Music From Text (Google Research, 2023), treat language as a control signal over generated music. That is industry context for why a text brief can steer generation. It does not describe this product’s internals.
Pipeline A — split lyrics from style
Open Text to Music when the brief is verbal.
Custom is the mode for locked lyrics: put the words in the lyrics field and the musical direction in the style field. In style, name genre, mood, instruments, pace, or the role the music will play in a video or song. You can create a song that uses those lyrics, or an instrumental if you want the style without sung words. This is also the path for making music from lyrics you already wrote.
Inspiration is the mode for a single whole-song prompt. Use it for background music for a video when the brief is one descriptive sentence, or for instrumental music without lyrics when you would rather not invent a lyric field.
Pipeline B — start from a still when the picture is the brief
Open Photo to Music when a visual is already doing the briefing work:
- Choose an image — a photo, an illustration, a cover image, or a video still.
- Add an optional written music note (genre, pace, instruments, or story) only when the image leaves something unsaid.
- Create a song or an instrumental.
The written note is optional. Typical jobs: original music for a video scene taken from a still; turning artwork or cover art into a song idea; music for a photo post.
Jade Copet et al., Simple and Controllable Music Generation (MusicGen / Meta AI, NeurIPS 2023), discuss generation conditioned on a textual description and, separately, on other controls. That paper is useful context for why splitting lyrics from style, or pairing an image with an optional note, is a control-design choice. It is not a claim that this product uses that model, and its scores are not product metrics.
Anti-patterns
- Pasting finished lyrics into a single Inspiration prompt so style language and verses fight for the same sentence.
- Inventing lyrics because a Custom form has a lyrics field, when you actually wanted an instrumental.
- Writing a long note that only restates what the still already shows.
- Treating a research paper’s numbers as if they belonged to the button you clicked.
Ship when the filename records the control (custom-lyrics, inspiration, or photo+note / photo-only) and you have not mixed three briefs in one pass.
Top comments (0)