Use case
Audiobooks with the atmosphere already inside.
Plain TTS reads the words. Seed-Audio 1.0 performs the page: the narrator’s pacing, a character’s voice inside a quote, and the waves under a lighthouse chapter, rendered together in one pass. No separate ambience track, no mixing session.
Hear it on one paragraph
This take is a single prompt: narrator description, the passage, one direction about the last sentence, and the sound of the sea kept under the voice.
The prompt pattern that works
An audiobook narrator, warm middle aged male, reads: 'The lighthouse keeper counted the ships every night. Tonight, one light too many burned on the horizon.' He gives the last sentence a chill. Faint waves and wind under the voice.- Fix the narrator first. Age, texture, register. Reuse the exact same description in every chapter prompt so the voice stays consistent across takes.
- Direct the reading, not just the text. One instruction per emotional beat ("he gives the last sentence a chill") beats adjectives sprinkled everywhere.
- Keep the bed under the voice. Name the ambience and tell the model to keep it low. For dialogue-heavy fiction, describe each character voice once.
Honest limits for audiobook work
- Takes cap at 2 minutes, and voiced text reads best under about 400 characters per take. Production means chaining takes; the consistent-narrator trick above is what makes chains seamless.
- English and Chinese only today.
- For a 10-hour plain-narration audiobook with no atmosphere, a long-form TTS platform is cheaper and simpler. Seed-Audio earns its keep when the book deserves a soundtrack: fiction, kids’ stories, immersive nonfiction.
Try it on your own page
Paste a paragraph of your manuscript into the playground, describe your narrator, and listen. The prompt grammar guide covers the cue sheet structure in depth.