Short-song pipelines often collapse “finish the track” into one afternoon: raise the loudness of a file that is not mixed yet, or split a mix that only needed to sit next to a playlist. This how-to keeps two lanes named. A workspace like WAV AI helps when you pick the lane before you click.
Decide the lane from the deliverable
| Deliverable | Lane | Do not open yet |
|---|---|---|
| Stereo file that should sit next to a release, video, or playlist | Master the finished mix against a different reference | A stem split “to see the parts” |
| Instrumental, isolated vocal, or one instrument to study | Split the mix | A mastering pass on a file whose parts are still the real ask |
| Arrangement or balance still changing | Neither — finish the mix first | Both in the same undifferentiated click |
iZotope’s explainer “What is mastering?” is the shortest third-party reminder of the order: mastering comes after the mix. Mixing balances separate tracks; mastering listens to the whole mixdown and polishes that stereo file for consistent playback and distribution.
Pipeline A — reference-guided mastering
Use AI Mastering only when arrangement and mix are already ready:
- Choose the finished song (upload it or pick it from the Music Library).
- Choose a different reference track that already has the loudness and overall sound you want to move closer to. The same song cannot be both the target and the reference.
- Create the mastered version, then compare original vs mastered.
The reference guides loudness and overall balance. Typical jobs: prepare a song for release; match music that sounds quieter or less even than surrounding tracks in a video or playlist; compare two final directions by running the same song against two different references. Export with a date and keep the unmastered mix for rollback.
Pipeline B — split when the parts are the product
Open Stem Splitter when the stereo master is not the deliverable:
- Choose a song (upload or Music Library).
- Choose the split.
- Listen to each part before you commit it to a rehearsal, karaoke bed, or edit.
Vocal remover produces an instrumental version. A 4-stem split gives vocals, drums, bass, and other elements. A 6-stem split also separates guitar and piano. Those part names overlap the four-family convention (vocals / drums / bass / other) used in source-separation research such as Alexandre Défossez’s Hybrid Spectrogram and Waveform Source Separation (Demucs / MDX 2021). That paper is industry context for why splitting a mix is a different job from finishing a stereo master. It is not a description of this product’s internals, and its quality scores are not product metrics.
Anti-patterns
- Mastering a draft whose arrangement will change tomorrow.
- Using the song as its own reference.
- Splitting a finished mix only because it sounded small beside other tracks—try a reference-guided master first.
- Treating a research paper’s numbers as if they belonged to the tool you clicked.
Ship when the filename encodes the lane (master-ref-A or stems-4) and the unprocessed mix is still one hop away.

Top comments (0)