Kokoro 82M

A small, fast, inexpensive model with thirty-three preset voices, labelled by accent (American, British and others) rather than by language, because it is an English model rather than a multilingual one. If your work is in English and there is a lot of it, this is the economical choice.

It supports a speed control but no emotion selector and no language picker. As with any model in that shape, the direction has to come from the writing: how a line is punctuated and broken determines how it reads.

Use it for scratch tracks, for placeholder narration while a cut is being assembled, and for finished work where a clean neutral read is all the piece needs. Move to MiniMax Speech 2.8 HD when a line needs a specific emotional delivery, or to Gemini 3.1 Flash TTS when the language is not English. Neither is something this model is trying to do, which is why it costs what it does.

Describe what you want in your own words and Kuro writes the prompt for you. If you would rather write it yourself, this is what Kokoro 82M responds to.

A small, fast English model with thirty-three preset voices, labelled by accent — American, British and others. af_bella is the default.

It has a speed control and nothing else to direct it, so the writing is the direction: how a line is punctuated and broken decides how it reads. A mood written into the text is spoken, not performed.

It is the right call for drafts and rehearsal — scratch tracks, placeholder narration while a cut is assembled — and for finished work where a clean neutral read is all the piece needs.

What it can't do

• Preset voices only; no custom voices.
• No style direction, no emotion and no language selector. The voice is the language, and it is an English model.
• At most 4,000 characters per take. Output is WAV.

Use something else when…

• A line needs a specific emotional delivery: MiniMax Speech 2.8 HD.
• The narration is in a language other than English: Gemini 3.1 Flash TTS.
• The delivery has to turn mid-line: Eleven v4.