2026-09-23 Changes confirmed medium confidence

Google Launches Gemini 3.8 Flash TTS With Custom Voice Design

The two speech-generation models add prompt-built voices, line-level performance control and scaled dubbing, while access and voice replication vary by product and region.

Edited by Tyronne Panaino

Google introduced Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS on September 23, adding two speech-generation models to its Gemini Audio family. Developers can use both through the Gemini API and Google AI Studio, while the consumer and enterprise rollout differs by model and surface.

The release matters to teams building voice agents, dubbing systems and produced audio because Google is moving beyond a fixed set of preset voices. Its announcement describes prompt-based voice design, line-by-line direction and reusable voice profiles, but it does not provide independent evidence for output quality, production reliability or the effectiveness of its safeguards.

A speech-generation layer alongside Gemini Live

Google positions the new TTS models as complements to its existing Live, transcription and translation models. Gemini 3.8 Flash TTS is aimed at deeper creative direction and character design. Flash-Lite is positioned for higher-volume dubbing, audio production and expressive agent speech where cost and scale are more important.

That distinction gives buyers a clearer product choice than a single speech endpoint. Flash is the model for designing and directing a performance; Flash-Lite is the volume-oriented option. The announcement does not publish independent latency, cost or reliability comparisons, so those remain workload-specific questions rather than established advantages.

Voice design becomes a programmable workflow

For Flash TTS, Google says users can define a role, accent and voice characteristics through natural-language prompts across more than 100 languages and dialects. The company also describes a library of more than 2,000 production-ready voices and the ability to reproduce a consistent vocal profile from a 30-second reference sample when the user has the right to use it.

Both models support line-level direction for pacing, emotion and conversational cues. Google also lists long-form generation, native two-speaker scene staging and scripted vocal sounds as supported production controls. Saved custom voices are intended to reduce drift across an ongoing project, while a separate voice-remixing feature is still described as coming soon and should not be treated as generally available.

Consent controls do not remove deployment risk

Google says voice replication requires a verbal consent recording from the voice owner that matches the reference speaker. It also says generated audio carries SynthID watermarking and that replicated voices use C2PA credentials. Those are concrete design controls, but this first-party release does not independently test whether consent checks resist impersonation attempts or whether downstream platforms preserve provenance information.

Availability also has a material boundary. Voice replication through AI Studio is not available in Illinois, Texas, the European Economic Area, the United Kingdom, Switzerland or India. Teams serving those locations therefore need to test the exact product path rather than assuming every capability in the announcement is available worldwide.

Access differs across products

Gemini 3.8 Flash TTS is rolling out to developers in the Gemini API and Google AI Studio and to general users through Gemini Notebook. Flash-Lite is also rolling out through the developer surfaces and appears in Google Vids. For enterprise customers, API availability in Gemini Enterprise is still listed as coming soon for both models.

The next useful checkpoints are published API pricing, measured latency and reliability under production load, wider enterprise availability, and independent evaluation of voice consistency and consent enforcement. Until those arrive, the supported conclusion is about the feature set and rollout boundaries, not superior performance.

Status

Confirmed. Google published the model announcement and current availability. Internal confidence is medium because all feature, safeguard and rollout details come from Google and were not independently reproduced in this run.