Gemini 3.8 Flash TTS Launches With Voice Cloning and 2,000 Voices

What did Google launch with Gemini 3.8 Flash TTS?

Gemini 3.8 Flash TTS is Google’s new expressive text-to-speech model for generating directed AI speech, including character voices and scripted scenes. Google introduced Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS on September 23, 2026, describing Flash TTS as its model for detailed creative control and Flash-Lite TTS as its model for high-volume audio tasks. Google’s launch announcement sets out the different roles for the 2 models.

Gemini 3.8 Flash TTS is aimed at work where a voice needs a distinct performance, such as a fictional character, a narrated scene, or dialogue between 2 speakers. Flash-Lite TTS is intended for high-volume dubbing, audio creation, and voice-agent work. The distinction matters because a project that needs creative direction has different requirements from a project that needs to produce a large number of consistent audio clips.

Google’s rollout also extends the company’s audio tools beyond basic read-aloud output. Readers who have followed Gemini’s Notebook audio features should treat the new TTS model as a separate voice-generation capability rather than an update to live voice chats or study tools.

How does Gemini 3.8 Flash TTS control a generated voice?

Gemini 3.8 Flash TTS accepts natural-language prompts for voice generation, which means creators can describe the intended delivery instead of relying only on a fixed synthetic voice. Google says Flash TTS supports line-by-line direction for acting cues, pacing, dialect shifts, conversational backchanneling, and native 2-speaker scripted scenes.

Gemini 3.8 Flash TTS can generate voices across more than 100 languages and dialects, according to Google. That range can help teams create localized audio without treating every language version as an identical reading of the same script. The limitation is that creative direction does not remove the need for human review, particularly when tone, pronunciation, or dialogue pacing matters to the intended audience.

Google also says its voice library contains more than 2,000 production-ready voices. A large library gives creators more starting points before they consider a replicated voice, which can reduce the need to use a real person’s vocal identity for an ordinary project. The practical approach is to use a library voice when a distinct individual’s identity is not necessary.

What is the difference between Flash TTS and Flash-Lite TTS?

Gemini 3.8 Flash TTS prioritizes expressive direction, while Gemini 3.8 Flash-Lite TTS is positioned for high-volume audio generation and voice-agent work. Google announced both models together, but the company describes their intended uses differently. Choosing the model based on the audio task matters because creative voice work and scaled production place different demands on a speech-generation system.

Gemini 3.8 Flash TTS is the more relevant option when each line needs an intentional performance. Gemini 3.8 Flash-Lite TTS is the more relevant option when the central requirement is production volume. Google has not presented the 2 models as interchangeable, so developers should test the model that matches their actual workflow before building around it.

How does Gemini 3.8 Flash TTS voice replication work?

Gemini 3.8 Flash TTS can create a voice profile from a 30-second voice sample, but Google requires a spoken consent recording from the voice owner that matches the reference speaker. The consent requirement is the central safeguard in Google’s voice-replication design because a voice can function as a recognizable part of a person’s identity in calls, recordings, and other audio interactions.

Voice replication is useful when a speaker has approved a project and needs a generated version of their own voice. Voice replication is not an appropriate shortcut for copying a colleague, public figure, family member, or client without their permission. The safe action is to obtain the required recording directly from the voice owner and preserve clear records of what the person agreed to create.

Voice replication also deserves a higher level of caution than selecting a generic synthetic voice. Recent security cases involving fake AI services and malicious impersonation show why users should verify unexpected audio requests through a separate trusted channel. A familiar-sounding voice alone is not proof that a message is authentic.

Where is Gemini 3.8 Flash TTS available?

Gemini 3.8 Flash TTS began rolling out in the Gemini API, Google AI Studio, and Gemini Notebook on September 23, 2026. Gemini 3.8 Flash-Lite TTS began rolling out in the API, AI Studio, and Google Vids. The rollout wording matters because access can appear at different times across Google products and accounts.

The generally available model IDs are gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts, according to the Gemini API release notes. The release notes also list a new /v1beta/voices endpoint. Developers should use the published model IDs and consult the API documentation before moving an existing workflow to the new voice models.

Google says AI Studio voice replication is unavailable in Illinois, Texas, the EEA, the UK, Switzerland, and India. That restriction applies specifically to AI Studio voice replication, not necessarily to every non-replication TTS feature or every Google product. Users who work across regions should confirm access in their own account before planning a production workflow around replicated voices.

What safeguards does Google apply to generated audio?

Gemini Audio output carries an imperceptible SynthID watermark, according to Google, and the company says C2PA credentials are used for voice-replication safeguards. SynthID is Google’s system for embedding provenance information into AI-generated media, while C2PA credentials are intended to provide information about media origin and related handling.

These measures matter because generated audio can be difficult to identify by listening alone, especially when a voice is designed to sound natural or familiar. Watermarking and credentials can support transparency, but they do not replace careful review of the source, consent status, and intended audience for an audio clip. Organizations should establish approval procedures before publishing generated speech under a brand or a real person’s name.

Google’s approach reflects a broader push to set technical boundaries around AI-generated media. The company’s separate work on private AI data handling also shows why users should evaluate both an AI feature’s capabilities and the controls around the data used to operate it. The most sensible approach is to treat voice samples and consent recordings as sensitive project materials.

Who should use Gemini 3.8 Flash TTS?

Gemini 3.8 Flash TTS is most relevant for developers, media teams, educators, and creators who need directed speech rather than a standard text reader. Google specifically positions the model for character creation and detailed creative direction, which makes it suitable for projects where delivery, pacing, and dialogue structure affect the final result.

Gemini 3.8 Flash-Lite TTS is more appropriate for teams producing large amounts of dubbed content, automated audio, or voice-agent responses. The model choice should follow the work requirement, not the assumption that a more expressive model is always better. A high-volume task may need predictable throughput, while a scripted story may need more control over the performance of each line.

Gemini 3.8 Flash TTS also requires a clear internal policy when voice replication is involved. Organizations should limit access to authorized staff, document consent, and review generated clips before release. Stop and contact Google support or your organization’s legal and privacy team if consent is unclear, a voice owner withdraws permission, or a regional availability restriction affects a planned deployment.

FAQ

Is Gemini 3.8 Flash TTS available now?

Gemini 3.8 Flash TTS began rolling out on September 23, 2026, through the Gemini API, Google AI Studio, and Gemini Notebook. Access can vary during a rollout, so users should check the relevant Google product or API account.

Can Gemini 3.8 Flash TTS clone a voice?

Gemini 3.8 Flash TTS can create a voice profile from a 30-second sample when the voice owner provides a matching spoken consent recording. Google’s consent requirement means users should not attempt to replicate another person’s voice without direct authorization.

How many voices does Gemini 3.8 Flash TTS include?

Gemini 3.8 Flash TTS has access to a library of more than 2,000 production-ready voices, according to Google. Google also says the model can generate voices from natural-language prompts across more than 100 languages and dialects.

Is Gemini 3.8 Flash TTS voice replication available in the United States?

Gemini 3.8 Flash TTS voice replication is unavailable in Google AI Studio in Illinois and Texas. Google’s stated restriction is specific to AI Studio voice replication, so users should verify feature access in their own region and product.

Does Gemini 3.8 Flash TTS watermark generated audio?

Gemini Audio models apply an imperceptible SynthID watermark to generated audio, according to Google. Google also says C2PA credentials support its voice-replication safeguards, although users should still verify consent and source information before publishing audio.