AI Stems for a New Era of Social Media Music

Scroll through TikTok or Instagram Reels for five minutes and almost nothing is silent. Estimates put the share of social video content that includes music somewhere north of 80%. Sound has become social media's default setting.

That shift has rewired how people find music. A generation of listeners now discovers songs through a 15-second clip before they ever hear the full track on a streaming platform. The clip comes first; the "real" song comes second, if at all. Music and social media are intimately woven together now, feeding each other's growth in a loop that barely existed a decade ago.

---

The same song, a thousand times over

For anyone on the rights and metadata side, this shift drives proliferation. A single composition, or what we understand as one ISWC, no longer maps to one or two definitive recordings.

It maps to dozens, sometimes hundreds, of versions such as slowed and reverbed edits, sped-up cuts, bass-boosted remixes, lo-fi flips or extended intros. If registered, each variant would get its own ISRC. This is one of different factors potentially explaining why ISRC volumes are expanding so aggressively year over year.

AI-generated and AI-assisted music is accelerating this too, adding another layer of version proliferation worth its own conversation. More on this here.

Fragmentation is a structure to work within

The instinct has often been to treat this explosion of versions as noise: too many ISRCs, too hard to track. But if a single recording is really a bundle of separable, re-monetizable parts, the smart move is to build tools that work with that fragmentation.

That's the gap our AI Stems module closes. It separates an existing recording into its individual components: vocals, drums, bass, instrumentation ... cleanly and automatically, without needing the original session files.

Each stem can then be cataloged, licensed, and monetized independently. A producer needing an isolated vocal doesn't need the full multitrack; a sync team repurposing an instrumental bed doesn't need to clear the whole song.

If the industry already treats one song as many versions, AI Stems lets you treat one recording as many usable parts, unlocking value previously locked inside a single master. For catalog owners with deep back catalogs, that's a new revenue layer, available as an add-on to the platform you already use to manage your rights.

If the industry already treats one song as many versions, AI Stems lets you treat one recording as many usable parts.

The big picture

Social media has changed what a "song" can be. A new unit of consumption is now a clip, a version, a loop remixed until the line between an original and its derivative blurs. ISRC volume is exploding because music itself is fragmenting in real time.

The catalogs that win won't be the ones forcing music back into neat packages, they'll be the ones working at the same granularity the market has moved to, down to the stem level.

That's the level of detail AI Stems was built for.