Google's Gemini 3.8 TTS lets you design a character voice from a text prompt

Google's Gemini 3.8 Flash TTS turns a written brief into a reusable, directable character voice, but voice cloning is unavailable in the UK, EEA and several US states, and launch pricing doubles on 1 January.

Google released Gemini 3.8 Flash TTS and Flash-Lite TTS on 23 September, and the feature that matters for filmmakers is voice design: you type a brief like the demo's "woman in her 30's, pumped, emphatic with Brooklyn accent" and the model builds that voice, which you can save and reuse across a project. Scripts take line-by-line stage directions, two speakers per scene and inline cues such as laughs, sighs and "mhm" interjections, so a whole dialogue scene can be performed from one script. In practice, temp dialogue for an animatic or a solo short no longer waits on a booked voice actor. Launch pricing for Flash is $0.50 per million text tokens in and $9 per million audio tokens out (tokens are the units a model reads and writes), and both figures double on 1 January 2027. There are two catches, both in the fine print. Voice replication, which clones a voice from a 30-second sample after a spoken consent check, is unavailable in AI Studio in the UK, EEA, Switzerland, India, Illinois and Texas. And the ranking Google cites comes from Hume AI's Voice Design Benchmark, while launch post co-author Alan Cowen founded Hume and ran it until January. Nobody independent has compared it with ElevenLabs yet.

Sources: Google DeepMind launch post and Google developer pricing page, 23 September 2026.