Euphona's stem separation isn't one split that gets recombined - it's four separate models: 2 stems (vocals / instrumental), 4 (vocals, drums, bass, other), 5 (adds piano) and 6 (adds guitar). Stems come back at the source sample rate, up to 192 kHz, 24-bit or 32-bit float.

Our own advice on the site is to use the fewest stems that answer your question: two if you just need an instrumental, six only if there's a real guitar part worth isolating.

I'd like to know whether that matches how you actually work. When you pull a track apart, what are you after - a clean acapella, a drum bus to re-balance, something else?

Ten free minutes, no card, if you'd rather test it than answer: https://euphonaai.com/?utm_sourc...

Next on my list after the launch is a short before/after clip for the page, so you can judge the engine instead of reading about it.

While I cut it, a question for anyone who records or mixes for a living. Euphona runs mastering, stem separation, restoration, vocal tuning, mixing and spatial audio off a single analysis pass, so each step knows what the last one changed.

In your own workflow, which of those would you happily automate, and which one do you want to keep your hands on?

Ten free minutes, no card, if you'd rather try it than answer: https://euphonaai.com/?utm_sourc...

We launched Euphona here on September 16 and finished the day with 1 upvote, 1 follower and no comments. Not the number I wanted, but the honest one.

What we built: one AI engine that analyzes a track once, then handles mastering, stem separation, restoration, vocal tuning, mixing and spatial audio, so each step knows what the last one changed. Every result can be compared level-matched against your original.

What I got wrong: I launched without an audience already here, and with screenshots for a product you have to hear.

What's next: a short before/after video on the page, and a first round of feedback from people who record for a living. Ten free minutes, no card: euphonaai.com