Splitting a completed stereo mix back into its component parts was impractical for decades and is now routine. The consequences reach further than remixing.
What separation actually does
A separation model is trained on recordings where the isolated parts and the finished mix are both available, learning to map one to the other.
Applied to a mix it has never seen, it estimates what each component would have been, producing approximations rather than the original recordings.
Quality varies by source. Vocals and drums separate cleanly because they occupy distinct spectral and rhythmic space, while similar instruments layered together separate poorly.
Why the artefacts matter less than expected
Separated audio carries characteristic smearing and residual traces of the parts that were removed.
These are audible in isolation and largely disappear once the part is placed back into a dense mix, where other material masks them.
Which is why separation is acceptable for production use and unacceptable for archival release, and the distinction is well understood by the people using it.
What it opened up
Recordings whose multitrack tapes were lost, degraded or never kept became workable again, which covers a large portion of everything recorded before the digital era.
Live performance capture, radio broadcasts and rehearsal recordings can now be cleaned and rebalanced rather than being unusable.
Restoration work on historical material has expanded considerably for this reason, since the source no longer has to be the master tape.
How the constraint moved to permission
Making an isolated vocal from a commercial release is now trivial, and using it in a released work remains a licensing matter.
Sample clearance was always the real barrier, and separation removed the technical obstacle while leaving the legal one exactly where it was.
The visible result is an enormous volume of unofficial remix work circulating outside commercial release, and a licensing system that was not designed for that volume of requests.
Where it changed live performance
Backing tracks can now be adjusted on the night, with a part removed for a guest musician or a vocal isolated for a stripped-back arrangement.
Karaoke, teaching and rehearsal all benefit from the same capability, since removing one part from any recording no longer requires a specially prepared version.
The tool that was built for remixers has ended up used most often by people learning an instrument, which is a common pattern for capabilities that lower a technical barrier.