"Scottish accent, tired, mildly annoyed" — that's all it takes for Gemini 3.8 Flash TTS (Text-to-Speech) to build a matching AI voice. No casting call, no recording booth. Just a text field.
Google unveiled the new model on September 23, alongside the leaner Flash-Lite TTS variant. Both are now available through the Gemini API and in Google AI Studio.
From 30 to over 2,000 voices
Gemini used to ship with around 30 preset voices. Flash TTS turns that into a library of more than 2,000 — not a fixed list to click through, but voices you shape with plain language. You describe a role, an accent, a character, and the model builds a voice around it. Over 100 languages and dialects are covered, down to regional varieties like Mexican Spanish, Quebec French, or Scots English.
The core idea isn't new — ElevenLabs and others have offered synthetic voices for a while, and decent AI narrators have been standard in audiobooks and ads for years. What's new is how deeply Google is now baking this into its own ecosystem: AI Studio, the Gemini API, and eventually Gemini Enterprise.
Two models, two jobs
Flash TTS is built for creative control: game characters, elaborately staged audiobooks, multi-voice podcast production. Flash-Lite TTS is the cheaper, faster variant meant for high-volume work — dubbing, automated narration, voice agents that talk to you on the phone.
If you're just curious, the distinction barely matters — both sit behind the same interface. It only matters once you're narrating thousands of lines a day.
What does it cost?
Both models are currently free to try in AI Studio, with the usual free-tier limits every AI provider imposes: a capped quota, no guarantee of constant availability. Using the models in production through the API is billed by text and generated audio length. As of now, that works out to roughly 1.35 US cents per minute of generated speech for Flash TTS, and about 0.9 cents for Flash-Lite TTS. Those are introductory rates, though: Google's own price list has them doubling on 1 January 2027.
Nothing new here: free to experiment, and regular use eventually costs money. That's not a scandal — it's just how data centers get paid for.
Try it yourself
A free Google account is enough to get started in AI Studio — no subscription, no credit card required. If you've already used the Gemini API, you can jump straight in; everyone else gets a web interface that works without writing a single line of code. Type some text, describe a voice in a sentence, hit play — done. You only need developer skills once you want to wire the API into something else.
One caveat, though: the better AI voices get, the easier they are to abuse for fake audio — voice-cloning phone scams are no longer science fiction. Google promises watermarking technology (similar to how it's done with AI images) and usage rules against abuse. Whether that holds up in practice remains to be seen — but that's not a reason to write off the tool itself, any more than it is for image generators.
Is this for me?
To play around with: yes, for anyone. For actual production use, it's more relevant once you have a concrete project — a podcast, an explainer video, a small game, a voice-automation workflow. If you just want to hear a weird AI voice once, the free quota in AI Studio covers you fine. For anything that runs continuously, check the actual API prices first — they tend to shift faster than anyone would like, with pretty much every provider.