Announcing UnMixIt Pro: The Stem Separator You’ve Been Waiting For
by UnMixItTeam   5 days ago

The separation I’m waiting for is one that nulls when you combine the various stems and test it against the original file. Does this do that?

Giving it a try is the best way to confirm it meets your specific needs—and since it's free, there's no risk in testing it out. We’re always here for feedback, too!

You don’t know if it nulls?

Okay…I’ll take that as a no. Good luck with your release but that’s what I am waiting for.

That was a fair question, and I'm surprised the dev didn't respond with a definitive.

Deflecting the very first question in the thread does not inspire confidence.

What's the price
What's the copy protection

Product page is more marketing, no price, no licence information. Clicking on "get started" (whatever that might mean) tells me I need to sign in with or create a Google account to continue.

So much sketch.

I'm interested, but I'll come back later to see if the developer (or vendor) treats customers with a more straight forward approach that is not a lesson in "marketing focused lead acquisition".

It looks like it is one of those AI subscription services where you upload your content to their service.

I'll pass.

https://digitalmagicwand.com/buy

For the record, SpectraLayers Pro is the only solution I’ve tried that accurately nulls when all stems are re-summed. It’s not inexpensive by any measure (but in fairness, it does a whole lot more than stem separation); nonetheless, I feel it sets the bar against which new tools like this should be measured (especially if they are calling themselves “The Gold Standard”).

How many instruments/types of instrumenmts are separated, and which ones?
"Other instruments too" is as vague as vague can get. Which ones?

I find more advertising claims than clear info and hard facts on the site, and insisting on a Google email login is preposterous from my view.

Have I mentioned that most people I know hate subscriptions as much as I do?

So you better be damn good with lots of stems separated per audio to be taken serious.

Oh that’s good to know I had not heard about that I’ll check it out

I thought perhaps some of the posters on this thread were being a bit harsh - until I actually clicked the link.
The most base, clickbatey, "renew your norton antivirus today" marketing trash website ive ever seen posted on here.

I think their target market is....not this forum.

"Not just for Karaoke"
"Real" reviews posted on the website:
"Im not a tech-savvy person"
"I've been messing around with music production and..."

On top of that they use a "credit" system that charges you per minute. I think its safe to say we arent the target audience

Aimed at the dev. Unparticipated.

That's the case with spectralayers 11 if I remember well.

Agree that SpectraLayers Pro is my Gold Standard and nulls down past -170db which suits my purposes. That isn't to say there isn't a bit of bleed between instruments or misinterpreting of instruments. SL11 is a bit better about that. It is sufficient for my light re-mixing. Sometimes using RIPX DAW or Lalal.ai or UVR may offer less bleed but those don't null as well as SL11 if that's what I need. It has been a somewhat expensive restoration journey no doubt.

Yes the bleed is unfortunately a byproduct of maintaining a “lossless” set of stems with a very good but ultimately imperfect separation algorithm. In practice I haven’t found this to be a major issue as the bleed tends to disappear in a mix or can be managed with gating/expansion (which usually yields a result very close to RipX in my experiments). SpectraLayers also has the “Unmix Drums” function on top of the “Unmix Song” feature, which lets me isolate kick, snare, etc. after creating a drum stem and still nulls all the way down the processing chain. This is a killer feature for my work and puts SpectraLayers over the top for me.

I don't care about the marketing. I just gave it a whirl and...

...I have to tell you guys, man to man. Man to woman...don't drop a product on the largest sound engineering forum in the world unless your product is ready for prime time. This place can be brutal...and I'm about to join the chorus after some first hand experience.

I had nothing but error messages from file size too large to no error message with nothing happening to the upload finally going through then alerting me of a 20 seconds limit even though I have the 4 free credits, 3 of which should work for a minute of audio.

After multiple attempts, I gave up.

I will pass for now and try out the true gold standard from a world renowned company, now that I've been made aware of it. Please don't sell my Google account info!

Sticking with SpectraLayers 11 Pro for now.

Which has a major advantage in it's a desktop app, source material stays safely in your control etc.

I swear, these guys uploading source material to cloud servers for processing just don't understand that ALL of the source material is kept for further training...reselling/sharing for ai model training etc etc etc etc.

Worse, the uploader likely has zero sr registration protection and will either never know the material is now "out there" unprotected, or worse...if ever realizes it "is" now out there....someone else has now preemptively registered an sr for it. Dumb dumb dumb dumb dumb.

Our recent benchmarks (see attached table) clearly show that UnMixIt Pro outperforms all other tools we've tested in stem isolation and recombination accuracy, including some of the latest commercial and academic systems.

While perfect nulling isn't yet achievable by any software, UnMixIt Pro comes very close, with minimal residual artifacts that are often below the perceptual threshold.

We’re happy to accept the challenge of an independent test. Just let us know if you are interested, and we can provide all the assistance you need.

Regarding pricing questions. We’ve designed our credit system to give you maximum flexibility across all our AI tools, ensuring you only pay for what you use. Here’s how it works:

Pro users: 3 credits per minute of audio.
Lite users: 1 credit per minute of audio.

Plus, for added peace of mind, we offer a 30-day money-back guarantee—no questions asked.

For full details on pricing and credits, check out our pricing page: https://digitalmagicwand.com/buy

If you have any questions, feel free to reach out—we’re happy to help!

Sorry to hear about the issues you ran into! We recently discovered a problem with certain FLAC file encodings causing upload errors, but it's already been fixed! Thanks for your patience while we worked on it.

To make it right, we’d love to offer you some extra free credits so you can give it another go.

And don’t worry, your Google account info is completely safe and never shared.

We hope you’ll give UnMixIt Pro™ another shot! Feel free to reach out if you need anything.

Hi electro,

Our pricing structure is straightforward:

Lite version: as low as $0.20 per song
Pro version: as low as $0.60 per song

(assumed on 2.5 minute average song duration)

Since UnMixIt is a cloud-based service, traditional copy protection doesn't apply.

The advantage is there's no setup required, no need for expensive GPUs, and you can use it anywhere - from mobile devices to PC/Mac!

We also offer a 30-day money-back guarantee, no questions asked!

For complete pricing details and a free trial, you can check our pricing page:

https://digitalmagicwand.com/buy

Let me know if you have any other questions!

Thanks for raising these concerns—they’re totally valid, and we’d be skeptical too! Here’s how we handle it:

1. Data Deletion & Privacy:
- Uploaded files are automatically deleted within 1 hour of processing. No exceptions.
- We do not train on user uploads, even if we wanted to. Here is why: Training requires original multitrack stems (which obviously you’re not providing), so it’s technically impossible to use your audio for training.

2. Why Cloud?
- Our earlier desktop version needed high-end GPUs (costing thousands) to match UnMixIt Pro’s quality—cost-prohibitive for most. Cloud processing democratizes access without compromising results.

We’re audio folks too, so we built this with the same paranoia we’d want for our own work. Happy to clarify further!

Apologies in advance to the developer (I'm sure you're good people, but)... honestly, utterly sick of these subscriptions, to the point where I just won't engage in a demo.

(and I don't tend to dish out harshness on forums!)

Ultimately, there's no way to know what a dev is/is not doing with uploaded source files.

Mixed source files that are uploaded by a user certainly CAN be demixed and/or further trained by a dev or whoever gathers the uploads. Uploaded files might be .....or might not ....be deleted.

It's likely the underlying code (based on the presence of 'other") is demucs or spleeter like everyone else. Maybe in-house tweaked.....maybe not.

If this supposed gold-standard code....and 55 nvidia gpus in the dev cloud.... are actually doing a gold standard, cutting edge thing ....like being able to pull a kazoo and 60 separate brass instruments from a recording of 1812 Overture onto separate tracks....I haven't read a claim.

Only structure offered is the standard fare that seems to indicate (to me) underlying demucs/spleeter. Nothing gold-standard in that.

Tweaking code to work quickly with desktop gpu (not cloud) takes some effort by a dev (SL for example).....but isn't rocket science....and...a dev committed to creating an app like this as a desktop app (not cloud) is TRULY understanding the concept of protection of source.....esp in these days.

This pile of code will ultimately be reviewed/compared/discussed by the usual github guys and neutral 3rd-party guys who do extensive test comparisons of ALL the available demix routines.

I'm not personally going to upload anything to this entity....the use of "gold standard" for a startup....a startup with zero pre-compare results to other commercial pgms already out there (Robin for example does comparisons NAMING each specific product with no bias).....is really off-putting to me.

At any rate, comparisons and results will find the light of day at some point by 3rd party testers.

It is so easy for devs to outline exactly what happens to users files, data and data derivatives in their t’s and c’s, yet this almost never seems to be the case.

Why? My assumption has always been that the files, data and data derivatives they glean from it are actually of more value to them in the long run than the product they are selling, and in some cases giving away for free.

The exchange isn’t necessarily a bad thing, except, IMO, with tools similar to this one l where a performers (vocalist, drummer, guitarist etc) art could be reused/remade and augmented/sold into other tools/services without their knowledge or consent and ultimately, no recourse.

As far as I can see, Unmixit Pro’s stated privacy notices absolutely does not go far enough to prevent post-deletion use of data or derivatives, especially in the context of AI.

The separation tool I’m waiting for is one that can recognize sounds that are not traditional ( guitars, vocals, drums, bass , piano, etc) , like a full electronic track with non-traditional sounds. Even Spectralyers struggle with that.