Design a distinct character voice from a text description alone

Why this works

ElevenLabs Voice Design generates a voice from the description itself rather than from a recorded sample, so the description is doing the entire job a casting call would normally do — a vivid one-line character sketch ("a weary lighthouse keeper who has seen one too many storms") reliably steers timbre and character more than any single structured attribute alone, which is why it leads the prompt rather than being buried after the demographics. The structured age/gender/accent/tone/pacing fields exist because Voice Design responds well to being given each axis explicitly instead of having to infer it from prose — leaving one out (especially pacing) is the most common reason a generated voice sounds generic rather than specific. Adding a short delivery context ("telling a ghost story to a child by candlelight") gives the model an implied emotional performance target, not just a static timbre, which is what separates a usable character voice from a flat text-to-speech read.