Creatune splits AI covers into Voice and Style tracks

The piece is framed as a way to organize judgment rather than a showcase of generated results, and Alex is explicit that it is not a measured report on any particular cover output. The stated premise is that you work only with audio you hold rights to and voices you have permission to use.

The method rests on separating decisions. A three-column table — keep, change, and leave untouched for now — is meant to make covers comparable, and Alex notes that the 'leave untouched' column is easily dismissed even though it is what keeps a comparison meaningful; making everything variable from the first attempt leaves no information to guide the next revision. Instructions for the voice are broken into musical roles rather than a person's name, covering range, attack, dynamics, texture and harmony, so the request can be reused as a production note. Style is likewise split into rhythm, instrumentation, density, space and the role of each section, with a worked example of dark rock described in concrete terms. Comparison is done on the same three segments of the original and the cover, and fixes move one column at a time, depending on whether the original character vanished, the change felt weak, or the results became incomparable.

What this hinges on is whether the separation actually survives contact with users' habits, since the tool itself allows either direction and neither forces the table or the permission checks. For creators the payoff would be revisions that teach them something, and for rights holders the risk sits in the step Alex places last: confirming the original song, the master, the voice and commercial-use terms for wherever the cover is published.