Write a voice-clone creation brief that captures the reference performance you actually need

Why this works

Voice cloning quality is decided almost entirely before generation happens, at the reference-recording stage, which is why this prompt is structured as a planning document rather than a direct generation request — a clone trained on 2-3 minutes of flat, single-register reading will only ever produce that one flat register back, so asking the AI to design a script covering a deliberate range of emotional tones and sentence lengths up front is what gives the eventual clone enough acoustic variety to sound natural across different real uses later. Including a written voice-attribute description as a fallback matters because recording conditions are often imperfect (a home office with no soundproofing, as in the example), and having a structured Voice Design prompt ready means a usable starting point exists even if the reference audio has noise or inconsistent mic distance. Making the consent requirement an explicit, non-optional output section rather than an assumed norm addresses a real and current failure mode in voice-clone workflows — people skip consent when it's not written down as a deliverable, and treating it as boilerplate the AI must produce alongside the technical brief keeps it from being quietly dropped when the project moves fast. The vocal-quirks field exists because generic clones tend toward a smoothed, characterless average voice; naming a specific quirk like a pre-point pause or a vocal rasp gives the source material something distinctive to preserve rather than average away.