The promise of an AI remix is simple: same song, new sound. Turn an acoustic demo into an EDM track, a pop song into a qawwali, a folk recording into lo-fi. The complaint is just as simple: the new version sounds great, but half the words are gone, a verse has turned into the chorus sung twice, or the singer is mumbling something that is not your lyrics.
This guide explains why that happens and how to avoid it, using the controls Autunes gives you. The short version: the dial matters more than the prompt, the lyrics matter more than the dial, and rap needs a different tool.
What an AI remix actually does
An AI remix does not edit your recording. It generates a new song from scratch while 'listening' to the original as a guide, and it sings the lyrics again with a new voice. The original shapes the structure, melody and timing; your style description shapes the sound; the lyrics tell the new singer what to sing.
That is how the Zori cover mode is designed: it re-generates the music guided by the reference, and it is not a note-for-note copy. That is the whole point of a remix, and it is also why words can go missing. The new singer is performing the song, not reading it back from the original track.
The dial: 'How close to the original'
Every remix and cover screen on Autunes has one dial, labelled 'How close to the original'. It sets how much of the render keeps listening to your original. It runs from 0.3 to 0.9, and since 25 September 2026 the number you see is exactly the value the engine uses.
What the positions mean in practice:
- 0.30 – 0.45, New take: the original is a starting point. Expect a different arrangement, a looser melody and lyrics that may drift. Good for 'inspired by' tracks, risky if you need the words.
- 0.45 – 0.65, Balanced: your style and the original share the song. The hook usually survives; dense verses may not.
- 0.65 – 0.85, Same song, new sound: melody and words stay, the production changes. The default, 0.75, sits here, and so does Studio's 'Cover' preset.
- 0.85 – 0.90, Almost the original: your style is a light tint on top.
Why 0.75 is the default
When we checked remixes that lost their verses, the pattern was consistent. At a mid setting the model hears the original for only part of the render and then continues on its own; sung hooks come through, but long or fast verses lose their timing and get replaced by the chorus. On a fast rap track in our tests, a mid setting kept only the hook, while around 0.7 both rap verses came back.
Our listening tests across genres point the same way: roughly 0.5 to 0.7 for a restyle where the song should stay recognisable, and 0.7 to 0.9 for a subtle change. So the default now starts at 0.75. Drag it lower only when you actually want a looser reinterpretation.
Step by step on Autunes
Remix works on songs you made on Autunes and on tracks you upload in Studio (Upload a track). For uploads you confirm that the recording is yours or that you have the rights to use it.
- Open the song. In Studio's song editor, pick 'Cover' (0.75, same melody, new style) or 'Remix' (0.5, a more creative reinterpretation). On the Create page, the Remix dialog and the Cover panel use the same dial.
- Describe the new style, not the old one. 'Acoustic unplugged, warm guitars, intimate male vocal' works; 'the same song but better' gives the model nothing to go on.
- Pick one main style and at most two modifiers. 'Lo-fi with a qawwali vocal and EDM drops' pulls three directions at once.
- Check the lyrics. If the lyrics box is empty, Autunes transcribes the original for you; if you know the real words, paste them, with [Verse] and [Chorus] tags. Transcription can mishear, and whatever is in the box is what gets sung.
- Leave the dial at 0.75 for your first try. Generate, listen, then move it: lower for more change, higher if lines are still missing.
Rap and very fast vocals: use Jam instead
Fast rap is where every AI remix struggles, ours included. Re-singing dense bars from text, in time, with a new voice, is much harder than re-singing a melody. If your song is mostly rap, the better route is Jam's 'My voice with music' option: Autunes lifts the original vocal out of the mix and builds a completely new band under it. The words cannot go missing, because they are never re-sung.
The trade-off is that the voice stays the same. If you want a new singer on a rap track, a remix around 0.75 to 0.85 with exact lyrics is your best shot, and it can take a couple of tries.
Rights: what you can remix
Songs you generate on Autunes are yours to remix as often as you like. Uploads are different: remix only recordings you made or have permission to use. A remix of a commercial release is still a derivative of someone else's composition and master, whatever tool made it. Voice cloning ('Sing it in my voice') is limited to songs made on Autunes for the same reason.