Gemini 3.8 TTS: Google Lets You Tweak Existing Synthetic Voices With a Simple Prompt
On November 4, Google rolled out Voice Remixing for Gemini 3.8 Flash TTS: starting from a library voice or a custom-designed one, users can adjust accent, pitch, tone, or speaking rate using a text prompt. The feature is restricted to paid projects and cannot be applied to cloned voices.
Google rolled out Voice Remixing on November 4, 2026, for Gemini 3.8 Flash TTS, its text-to-speech model launched in September. The core idea: rather than creating a voice from scratch, users can take an existing voice and modify it using a text prompt, such as “add a slight Southern US accent” or “make the voice a bit older, more grounded, with an audiobook cadence.” The feature is available in Google AI Studio and via the Gemini API.
How it works
Users start by picking a base voice: one of Google’s 30 studio voices, a voice from the expanded library (which Google says contains more than 2,000 voices), or an existing custom voice designed from a prompt (“Voice design”). From there, users describe the relative tweaks they want: age, pitch, energy, tone, accent, or speaking rate. The model generates a new, stored voice with its own unique identifier and a short preview sample, leaving the original voice untouched. This remix can then serve as the foundation for another remix, allowing creators to fine-tune a voice incrementally.
The resulting voices work with Gemini 3.8 Flash TTS and its lightweight counterpart, Flash-Lite TTS. The feature is restricted to paid API projects (Tier 1 or higher). A single project can store up to 200 custom voices across all categories, with each voice expiring one year after its last use.
What remains excluded: cloned voices
Gemini 3.8 TTS offers five methods to obtain a voice, including voice replication based on a recording of a real person. This approach comes with much stricter guardrails: Google requires a 10-to-30-second audio sample alongside a consent recording from the same adult speaker reading a prescribed script, and it rejects synthetic audio or files carrying SynthID or C2PA watermarks. According to the documentation, all synthesized audio carries an inaudible SynthID watermark.
Voice Remixing, meanwhile, is documented to work only on prebuilt voices, library voices, and designed or already-remixed voices within the project. Replicated voices do not appear in the list of supported bases: according to the documentation, users cannot take a cloned voice and run it through the remixing tool.
Our take
For creators working on videos, podcasts, or video games, Voice Remixing solves a very real headache: finding “almost” the right voice in a library, only to have to start all over again just to tweak one detail. Being able to hold onto a voice while dialing in an accent or pacing brings voice direction much closer to working with an actual voice actor. It is an incremental enhancement rather than a brand-new model, and its true quality will need to be judged through independent testing. The most notable takeaway is Google’s decision to restrict remixing strictly to synthetic voices while keeping cloned voices behind a consent verification flow: a sensible safeguard, even if it only covers what flows through Google’s own tools.
Sources
- Google (X) — Voice Remixing announcement for Gemini 3.8 Flash TTS Primary source
- Google AI for Developers — Voice remixing guide Primary source
- Google AI for Developers — Gemini API changelog (November 4, 2026) Primary source
- Google — Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS (September 23, 2026) Primary source
- Google AI for Developers — Speech generation guide (voice options) Primary source
- Google AI for Developers — Voice replication guide Primary source
A primary source comes directly from the company, lab or organization concerned. A secondary source reports the information without being its author: we flag it as such. Spotted an error? See our corrections policy.
