Simon Willison built a bring-your-own-key playground that renders single-voice narration and multi-speaker dialogue from a Gemini API key.
Google released two Gemini text-to-speech models on September 23, 2026: gemini-3.8-flash-tts and the cheaper gemini-3.8-flash-lite-tts. Both ship with a library of more than 2,000 voices and support a 30-second audio sample for custom voice creation.
Simon Willison published a companion playground that runs on a user’s own Gemini API key through the API’s open CORS policy. He wrote the interface with GPT-6 Astra and used Claude 4.5 Opus to script a demo conversation between two pelicans debating a move to the Pacifica Pier.
Builders get a single request format for multi-speaker narration, which suits audiobooks, training modules, and localized product demos. Voice style instructions let teams assign distinct delivery to each speaker without stitching separate audio files together.
Cost remains the variable to track: generating 78 seconds of audio with Gemini 3.8 Flash TTS took about 20 seconds and 2.74 cents, while the Flash-Lite tier is cheaper. Teams weighing volume production should benchmark both tiers against their own scripts before committing.
What matters
- Google released gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts with over 2,000 built-in voices.
- Builders can define multi-character conversations with per-speaker voices and styles in one request.
- Watch custom voice cloning, which needs a 30-second sample, plus the 2.74-cent cost for 78 seconds of audio.
Why it matters
Watch custom voice cloning, which needs a 30-second sample, plus the 2.74-cent cost for 78 seconds of audio.
This GenAI News article was prepared in original wording using reporting and materials published by Simon Willison’s Weblog. Source reference: https://simonwillison.net/2026/Sep/23/gemini-tts-playground/.
Drafted by the GenAI News review pipeline.