Every generator — Photo → 3D, Text → Image, Text → Speech and Text → Music — runs on Murakumo GPU generation through the same-origin /api/murakumo-generation proxy (ADR-0057/0058); no credential ever reaches the browser. Generated artifacts are content-addressed (CID) and become publishable assets in the Hub. The creative team (ADR-0028/0029) folds each picked persona's voice into the prompt before it's sent — a fictional creative persona citing one real creator's aesthetic lineage as inspiration, not that creator's actual work.