The core annoyance of generating a full song with AI has never really been the generation — it’s the editing. Get a verse you love and a chorus that falls flat, and most tools force a choice: regenerate the whole track and risk losing the verse too, or live with the chorus. ElevenLabs shipped a fix for that on August 25 with Composer, a new editing layer inside ElevenMusic that treats a song as a set of independently addressable sections rather than one indivisible generation.
Editing a section without touching the rest
Composer’s pitch is specific: open a track, select a verse, chorus, bridge, or outro, and regenerate just that section — adjusting its style, tempo, energy, instrumentation, key, or duration — while the rest of the composition stays exactly as it was. Lyrics can be added or edited per-section too, without triggering a full-song regeneration. The tool also supports generating multiple takes of a single section and auditioning them side by side before committing, and lets a writer restructure a song after the fact — swapping sections, doubling a chorus, tacking on an outro — the kind of arrangement decisions that used to mean starting over.
ElevenLabs hasn’t disclosed the underlying architecture change that makes section-level regeneration possible without disrupting the surrounding audio, but the practical effect is the thing worth noting: it moves ElevenMusic from “type a prompt, get a song” toward something closer to how a producer actually works a track — pass by pass, section by section, discarding what doesn’t work without discarding everything.
Why it fits here
AI music generation has mostly competed on how good a first pass sounds. Composer is a bet that the harder, more valuable problem is the second pass — and it’s a meaningful admission that “good enough on the first try” was never going to be the whole story for anyone actually trying to finish a song rather than sample a novelty. It’s live now for ElevenMusic users, not a preview or waitlist feature, and it landed the same week ElevenLabs also shipped a CLI tool and speaker-diarization writeup — a company visibly building out the unglamorous editing and tooling layer around its models rather than just the models themselves.