How do I choose a voice on a text-to-speech model?

Where the text goes

Text-to-speech models have no picture prompt: the prompt box is the script. Write exactly what should be spoken; LUVI sends it as the model's text input. Aspect ratio and negative prompt are hidden because they do not apply.

Picking the voice

Below the prompt, the parameters panel shows a Voice dropdown. On ElevenLabs v3 it lists 21 voices by name with a gender icon and accent, for example Roger (male, American), Sarah (female, American), Charlie (male, Australian); the default is Bella (female, American). Behind each name LUVI sends the provider's own voice id, so what you hear is exactly the vendor's voice of that name.

Gemini TTS and MiniMax Speech expose their own voice lists in the same place; Grok Speech has a smaller set.

Price

Speech is billed on the length of the text:

Model	Starting price
ElevenLabs v3	200 credits per 1,000 characters
Gemini 2.5 Flash TTS	80
Gemini 2.5 Pro TTS	160
Gemini 3.1 Flash TTS	300
MiniMax Speech	96
Grok Speech (xAI TTS)	30

Billing is linear with no rounding to blocks, and the text is measured after trimming spaces. The estimate under the Generate button follows your script as you type.

After generating

The result is an audio file in your project with a player. Use it as the audio reference for a lip-sync model, download it, or send it to a teammate. Through the Claude connector, the audio plays inline in the chat.