The Native Instruments Stems format launched at the 2015 Amsterdam Dance Event with a press release, a handful of partner labels, and a Traktor update that nobody outside the DJ press paid much attention to. The pitch was genuinely smart: ship a track and its four stems as a single file that still plays anywhere. For a few years it looked like it might become the way stems were bought and sold.

It did not. The format never reached the mainstream, the stores that sold Stems files have mostly stopped, and the premise it was built on is being overtaken by on-device stem separation. This is the story of a good idea that ran into a problem no file format could solve.

The arc of the .stem.mp4 format is a useful case study in why "ship stems as files" keeps stalling, no matter how well the file is designed.

What the format actually is

At the file level, the Native Instruments Stems format is an MP4 container with a specific structure. The MP4 spec already supported multiple audio tracks. Native Instruments wrote a convention on top of that.

  • Track 1: the original stereo master mix, exactly as you would hear it from a regular MP4. Plays in any media player. Backward compatible.
  • Tracks 2-5: four separately decodable stems, each in stereo, summing back to the master.
  • Metadata: per-stem names ("Drums", "Bass", "Vocals", "Melody", or whatever the engineer chose), color codes for visual coding in DJ software, and the usual MP4 metadata fields (artist, title, BPM, key).

The result was a file that would play correctly in any standard MP4 player, give DJs and remixers full stem access in software that understood the format, and stay under fifty megabytes for a typical track.

One of the smartest decisions Native Instruments made: they released the format as an open specification and did not try to license it. Any software vendor was free to read and write .stem.mp4 files. Any label was free to encode their catalogue without paying NI. That decision is the reason the format's playback support outlived its commercial momentum.

The 2015 launch and the slow start

Traktor 2.10, released alongside the format, supported Stems out of the box. NI signed launch partner labels (Spinnin', Monstercat, and several others) to encode catalogue tracks in the new format, and stores including Beatport and Traxsource started offering Stems files. The initial catalogue was a few thousand tracks, mostly EDM.

The first two years were quiet. Third-party DJ software (Serato, rekordbox, Virtual DJ) was slow to add support. DJs who were not on Traktor had no reason to care, and early on you also needed NI hardware to get the most out of it. Producers had no real workflow advantage from delivering Stems-encoded files versus just delivering raw stems to their label. The major-label catalogues stayed out of the format entirely.

There was also a real production cost. Encoding a track as a proper .stem.mp4 required the engineer to render four balanced stems that summed back to the original master. For most label catalogues, that meant going back to the multitrack session, doing the work, and re-mastering. For older catalogue, where the multitracks were missing or unusable, it was impossible.

The traction that never tipped

For a stretch around 2017 to 2019, it looked like the format might break through. Serato added Stems playback in 2019, finally putting NI's format inside the most-used DJ software in the world. The four-stems-in-one-file trick was genuinely useful for performance, letting a DJ mute the drums during a transition without any remixing ambition at all. Stores sold Stems files, and a catalogue slowly built up.

But it never tipped into the mainstream. The buyer base stayed niche, the catalogue stayed mostly EDM, and the major labels never came. Selling a stems file also asked DJs to re-buy music they already owned, just to get the stem version, which was a hard sell. By the early 2020s the format was widely described as stalled. Native Instruments stopped actively pushing it, and Traktor later shipped a conversion utility so users could make Stems files themselves, a tell that buying them had dried up.

Where it stands in 2026

The playback side of the format aged better than the commercial side. Stems files still play in most major DJ software:

  • Traktor Pro (the original, still maintained)
  • Serato DJ Pro (added support in 2019)
  • rekordbox (Pioneer DJ)
  • Engine DJ (Denon DJ, the InMusic ecosystem)
  • Algoriddim djay Pro
  • Virtual DJ
  • Mixxx (the open-source DJ software)

What has largely gone away is the buying. The stores that sold pre-encoded Stems files have mostly moved on, and there is no longer a healthy commercial catalogue to shop. The convention survives as something DJ software can read; the marketplace around it did not. Tellingly, most DJ software now leans on built-in real-time stem separation instead, splitting any loaded track into parts on the fly rather than relying on a pre-encoded file.

What the format got right

Three decisions still look smart in retrospect, and they are worth keeping even though the commercial format faded.

Backward compatibility. A .stem.mp4 plays as a normal music file in any standard player. The stem tracks are additive. A DJ who does not have Stems support still gets a usable file. That removed one of the adoption barriers most new formats die from.

Open specification. No royalty, no license, no NI veto on third-party adoption. Anyone could implement it without permission, which is why playback support is still everywhere long after the hype passed.

Standard container. MP4 is everywhere. Every audio engineer's tools support it. Every audio decoder library handles it. The format did not ask anyone to invent a new container. It just put a convention on top of one that already existed.

Where stems-as-files end

The thing the Stems format never solved, and was never going to solve, is the catalogue problem.

To encode a track as a proper .stem.mp4, you need the original multitrack session, an engineer to render balanced stems, a mastering pass, and a finished file. That is a per-track production cost. For new releases, it is a manageable line item. For the back catalogue, every recorded song ever made, it is prohibitive.

The catalogue problem is what has stalled every "ship stems as files" business model over the last fifteen years. There are not enough tracks. There never will be enough tracks. The universe of music a listener wants to remix is always going to be larger than the universe a label has rendered and shipped in a stems format.

The shift that has been accelerating since 2022 is the obvious one. Generate stems on demand, on the user's device, from any audio they have the rights to. Demucs, BS-Roformer, AudioShake's commercial models, and Apple's Logic Pro stem splitter have all crossed the quality threshold where on-device stem extraction is comparable to a properly engineered .stem.mp4, for a substantial fraction of recorded music.

Once a listener can extract clean stems from any track they are authorized to use, the value of pre-rendered Stems files mostly evaporates. That, more than anything, is why the pre-encoded Stems market never grew into the standard it could have been.

What it leaves behind

The Stems format will be remembered as a smart, honest attempt that proved the limits of its own model. Two parts of it deserve to outlive it: the idea that a stem file should be backward compatible and built on a standard container, and the open, royalty-free spec that let anyone implement it.

Our read at Remix.me is that the future is not a better stems file. It is stems that are produced when they are needed, from music people already have the right to use, with the earnings flowing back to whoever owns the track. Pre-rendered formats like Stems were the bridge between "stems are a studio asset" and "stems are something you can summon." A good bridge. Just not the destination.

Not bad for a 2015 ADE launch nobody outside the DJ press paid attention to.