A generation surface and an editing surface fail in different ways. The first fails when you do not have audio yet and still try to pick a specialized transform. The second fails when you already have a mix and you still wander an undifferentiated index. This technical note is about the two entry contracts in a product such as WAV AI, and the precondition each one assumes.
The catalog is an index with three groups
All Music Tools is titled All Music Tools. Its description is a single sentence: create original music, write lyrics, and reshape existing audio from one focused set of tools. That is an index contract, not an “open this file” contract.
The page materializes the contract as three groups:
- Music Tools: Text to Music, Photo to Music
- Lyrics Tools: Lyric Studio
- Editor Tools: Music Remix, Music Mashup, Music Extend, Music Replace, Stem Splitter, Music Mastering
Featured sections under those headings expand the same families. If the next action is still “which family?”, this is the right surface. If the next action is already “this file, this transform,” the index is a detour.
Jakob Nielsen’s Top 10 Information Architecture (IA) Mistakes (Nielsen Norman Group, 10 May 2009) includes “Missing Category Landing Pages”: a section overview helps people understand scope; linking only to leaf pages can hide what the section contains. I am citing that as information-architecture context for keeping a tools index, not as a usability score for this site.
The editor family assumes you already have audio
The AI Music Editor page states the precondition in the FAQ: yes, you need a song before you start; these tools work with music you already have. Inputs vary by tool: one song, two songs, an eligible generated song from the Music Library, or a finished song plus a different reference song.
The workspace is a family of six job-specific tools, not a single multi-purpose canvas: Music Remix, Music Mashup, Music Extend, Song Section Editor, Stem Splitter, and AI Music Mastering. The catalog card for the section-edit job is labeled Music Replace; the editor page calls it Song Section Editor. The page’s own rationale for separate tools is “make one clear change at a time.” You may chain tools on the same song (remix, then stem-split, then master with a reference), but each step is still one job.
That precondition is the important interface detail. An editor tool that wants an existing mix is a different function signature from a generator that wants a prompt or an image. Mixing the two signatures is how pipelines get noisy: you open Remix because it is nearby, when the actual problem is that you still need a first draft from Text to Music or Lyric Studio.
Yuancheng Wang and colleagues, in AUDIT: Audio Editing by Following Instructions with Latent Diffusion Models (2023), frame instruction-guided audio editing as a task that takes existing input audio plus an instruction, rather than a full target description from scratch. I am using that only as research context for the “input already exists” contract. It is not a claim that this product uses that model, and I am not quoting the paper’s scores as product quality.
A compact routing table
| Precondition | Surface | First action |
|---|---|---|
| No audio, or only a family (generate / lyrics / later edit) | Tools catalog | Pick the group, then the leaf tool |
| One existing track, new arrangement | Editor | Music Remix |
| Two existing tracks, one blend | Editor | Music Mashup |
| Track is right but too short | Editor | Music Extend |
| One eligible section needs new lyrics or style | Editor | Song Section Editor |
| Need isolated parts | Editor | Stem Splitter |
| Mix is finished; need a reference master | Editor | AI Music Mastering |
Stop when the note next to the file records the surface (catalog or editor) and the single job you ran. If you cannot fill the “input audio” column, you are still on the catalog side of the contract.
Top comments (0)