Covers, extend, repaint, accompaniment, stems
Work from existing audio: cover a song with any engine, extend or repaint part of a take, build a band around a vocal, split a take into stems.
Every task except a new song starts from a source: a WeavyAudio catalogue track (⤓ From catalogue), your own file (⬆ Upload, up to 60 MB, any common audio format) or one of your finished takes (the Further work buttons in its details).
Tasks
| Task | Engines | What it does |
|---|---|---|
| Cover | YuE2, HeartMuLa, ACE-Step | Keeps the melody (and chords/rhythm), sings your lyrics in a new style and voice |
| Reference voice | ACE-Step | Takes the voice colour and mix of the source |
| Extend | ACE-Step | Adds 5–90 s after the end; the last 4 s are re-rendered so the join has no seam |
| Repaint | ACE-Step | Re-renders one window (3–90 s); the rest stays as it was |
| Accompaniment | ACE-Step (base model) | Turns an a cappella vocal into a full arrangement with the instruments you tick |
Covers per engine
- YuE2 – SheetSage2 transcribes the source into an ABC score (melody + chord symbols, or melody only when Keep chords is off) and YuE2 sings that score with your lyrics. The score it used is in Details → Score.
- HeartMuLa – MuLaCover reads the melody, chords, drum pattern and tempo from the source and renders a new performance. Your style is sent as
topic / genre / instrument / mood. MuLaCover weights are CC BY-NC 4.0 – non-commercial. - ACE-Step – the source is the song plan (task
cover); Cover strength 1.0 follows it to the letter, lower values leave room for the new style and voice.
Lyrics are never taken from the audio: a catalogue pick fills them in, for an upload paste or write them (or use ✦ Write with AI).
Extend and repaint
Both keep your voice LoRA and settings. For extend leave the whole song in the lyrics and add the new lines at the end – the model needs the text to line up with the audio. For repaint pick the window with the two sliders and a mode: Careful stays close to the old part, Balanced (with a strength slider) is in between, From scratch is pure diffusion.
Stems and accompaniment
✂ Split into stems in a finished take's details separates vocals, drums, bass, guitar, piano and the rest (BS-Roformer for the vocal, Demucs 6-stem for the band). Each stem can be played and downloaded as MP3. Once a take has a vocal stem, 🎤 New backing for the vocal starts an Accompaniment task with that vocal as the source. You can also upload a dry vocal.
Accompaniment uses the ACE-Step XL base model (the only one trained for this task) – treat it as experimental.