While, for the past quarter of a century, we have specialised in recreating audio samples for use in new recordings, the world has changed dramatically over the past year. Today, a large proportion of our work involves recreating AI-generated vocals or even entire AI-generated songs using real musicians.
There are a number of reasons clients require this:
Legal: Some labels remain unwilling to sign AI-generated material because of concerns over copyright ownership, training data, or future legal uncertainty.
Commercial: Many important outlets, including numerous radio stations and other music platforms, are still unwilling to feature music that is entirely AI-generated.
Creative control: AI tools such as Suno can generate a complete song in seconds, but they offer relatively little control over the finer details. Recreating the track traditionally allows the artist to refine the arrangement, performance, lyrics, production and mix in ways that simply aren’t possible within the AI workflow.
Sound quality: At the time of writing, AI-generated music is still noticeably lower in fidelity than a well-produced traditional recording. The sound is often fuzzy, dense and lacking in transient detail, rather like comparing a low-bitrate MP3 with a CD or a high-quality vinyl pressing.
Ethical reasons: Occasionally, clients simply prefer that the final commercial release is performed by human musicians.
Unfortunately, unlike traditional sample recreations, where our goal is usually to produce an almost indistinguishable re-recording, that is not always possible with AI material. AI systems frequently hallucinate sounds that no real instrument produces, or generate vocal performances that no human singer could physically recreate. As a result, the objective is often not a perfect one-to-one copy, but rather to answer the question: “What would this song sound like if it had actually been performed by real musicians?”
One of the biggest AI viral hits of 2026 has been The Puerto Rico Song. While the track initially exploded online in its fully AI-generated form, we have since recreated the entire recording using live musicians, and the commercial release is now based on this new version.
In this particular case, the original AI production was already relatively close to a conventional recording, so although the live version is not identical, it is similar enough that most listeners probably wouldn’t notice the difference unless they were directly comparing the two.
Ironically, once a song has already become famous in its AI form, much of our work shifts to making it sound more like the AI original, rather than simply making it sound better. We often find ourselves deliberately blurring the sound, softening transients, smearing vocal textures and recreating some of AI’s distinctive artefacts, simply because those have become part of the sonic identity that audiences expect. In effect, we are making real musicians sound slightly more artificial in order to preserve the character of the original recording.
It will be fascinating to see how this kind of work develops over the coming years as the legal, technical and ethical aspects of AI music continue to evolve. For now, however, our role is a fairly straightforward one: employing talented human musicians to perform on records, just as they always have. Whatever direction AI ultimately takes, that’s one part of the process that still feels like an unequivocally positive thing.