Manuscript Studio

Narrators and voices

The narrator engines, which plans reach them, how they differ in performance control, and auditioning a voice before you spend anything.

The engines

Engine Plan Performance direction
Kokoro (included) Creator and above None
Fish Audio S2 Pro Premium and above Per-passage cues
Microsoft MAI-Voice-2 Flash Premium and above One direction per request
Gemini 3.1 Flash TTS Studio Per-passage cues

The Setup stage lists engines your plan cannot reach as well, marked unavailable — so an upgrade shows you what it buys rather than hiding it.

They differ in more than voice quality

The column that matters is performance direction, and it has three states, not two:

  • None — the engine narrates the text and nothing else. Cues are ignored. This is the included narrator.
  • One direction per request — the engine takes a single instruction for the whole passage ("weary, unhurried"), but cannot be told to whisper one clause.
  • Per-passage cues — direction can be placed cue by cue, inside the text.

"Supports direction" would conflate the last two, and they are genuinely different to work with.

Cost

The included narrator is cheaper than the others by a wide margin — on the order of a few tens of cents for a full book, against several dollars for an expressive engine.

That gap is why engines are banded by plan at all. It is also why the included narrator is the right default: for a lot of books, plain competent narration is what you want.

Auditioning

▶ Preview voice synthesises a fixed sample line in the voice you have selected and plays it back.

It costs nothing. Use it. Voice choice is the single most consequential decision in an audiobook, and the difference between two voices reading your genre is obvious in five seconds and invisible in a description.

Switching engines later

Direction is written once, in a neutral vocabulary, and compiled for whichever engine you use.

So switching engines does not mean re-tagging the book. It does mean re-rendering — the engine is part of a passage's identity, so changing it makes every passage stale, and every passage is charged again.

Decide the engine before you render a whole book, not after.

What to watch for

  • An engine your plan cannot reach is refused by the server, not silently swapped for the included one. Being quietly narrated in a voice you did not choose — and charged for it — would be much worse than an error.
  • The included narrator will ignore your direction entirely. It is not broken; it has no expressive control. If you have tagged a book for performance, you need an expressive engine to hear it.
  • Voice and engine are per book, not per account, so different books can use different narrators.