AI Music in 2025 – Notes on a Stalled Revolution
It’s been over a year since we wound down our generative music startup okio. Since then the space has “evolved” in ways both predictable and boring.
We now have 50+ high-quality video gen models, yet music still has just a handful (suno, udio, stability, 11labs…) - all closed source, all sounding somewhere between bland and absolute dog poo (unless you burn weeks on trial-and-error hacks).
The music industry’s lawyer army has strangled innovation, while many musicians have turned violently anti-AI. UX innovation has stalled too: most tools haven’t evolved in a year, others now mimic a shitty DAW in the browser.
Some sell “ethically sourced data” (usually half-true), but it's pointless as outputs still sound like dog poo. Google showed off real-time gen, but in mega-corp fashion it’ll never be a real product - they'll ship a muzak machine for video slop at scale.
The only glimmer: Chinese dark-horses might drop open-source, uncensored models trained on everything. They’ll sound great—and get demonized in the West.
Shame, because the promise of generative music has never been bigger.
Music is humanity’s most direct technology for emotion. And yet AI music has been reduced to slop and jingles. It’s a tragedy of imagination: the space where innovation should be loudest is dead quiet.
My bet is that the real breakthroughs won’t come from the lawyer-choked West or from UX gimmicks. They’ll come from the underground, open source, and global scenes - just like every real musical revolution before. Onwards!