Shape a voice. Generate the moment.
Clone a reference voice, design one from a description, or blend both — from a single sentence to a 30-minute narration. Long texts are split, generated chunk by chunk, and merged for you.
Voice generation console
Pick how to generate, then choose a voice.Speak naturally in a quiet room. The clip is transcribed automatically.
Compose speech
Write or paste anything — up to ~30 minutes of audioAdvanced options
How strictly the voice follows your Voice direction text. Low values sound more natural and relaxed; high values follow the description more literally but can sound exaggerated. Used in Direction and Design, and whenever a pace other than Natural is set (pace is sent as a voice instruction). Clone and Plain at natural pace always use 1.0.
Long texts are generated in pieces of about 12 s (Short), 28 s (Balanced) or 45 s (Long). Longer pieces mean fewer joins and a more continuous tone, but the delivery can slowly drift (slower, flatter) towards the end of a long piece — most noticeable with slow paces or repetitive text.
The starting point of the model's randomness. The same text, voice and seed give the same result every time; change it (or press the dice) to get a different take with the same settings. For long texts every chunk uses this seed.