Google's Gemini 3.8 Flash TTS Now Remixes Voices With Plain-English Instructions

Google has enabled Voice Remixing in Gemini 3.8 Flash TTS , delivering a feature listed as coming soon since the model’s September general-availability release. Developers can modify an existing voice with natural-language instructions, reducing the need to create a new persona or use a preset unchanged. The feature is available through the Gemini API and Google AI Studio.

A remix prompt can adjust timbre, the tonal character of a voice, along with pitch, pace, accent, and delivery. In Google AI Studio, developers choose a base voice, enter an instruction such as “soften the delivery,” audition three generated takes, and select one for further use.

Direct a voice you already like

The starting point can be a custom persona or any selection from Google’s library of more than 2,000 preset voices. Prompts can request regional characteristics, such as a subtle Southern U.S. accent, or stylized performances resembling a fire-breathing dragon.

Gemini already supported free-form prompts for designing voices. The third-party Hume AI Voice Design Benchmark ranks Gemini 3.8 Flash TTS first overall with a score of 71.4 and first for accent modeling with 60.8. Remixing applies that prompt-based control to an existing voice, allowing targeted changes while retaining the qualities that made the original useful.

Choose fidelity or throughput

Google’s current TTS lineup includes two sibling models. Flash emphasizes expressive fidelity, regional accents, multi-speaker dialogue, and long-form stability. Flash-Lite targets higher-volume applications such as real-time agents and read-aloud features. Google’s Voice Remixing announcement applies to Flash.

Hume AI’s Overall Quality Index ranks Flash first and Flash-Lite second. Those results support Google’s positioning of Flash as the creative model and Flash-Lite as the production-oriented option.

Tune a voice without rebuilding it

Production voices often match the intended character while missing a specific requirement, such as slower pacing or a more accurate regional accent. Previous workflows relied on accepting the preset, generating a new persona from a description, or replicating a voice from a 30-second recording. Remixing starts with a known voice and narrows the changes to named characteristics.

• Three-way audition: Google AI Studio generates three takes for each remix request.
• Prompt-level control: Instructions can target delivery, pitch, pace, timbre, or accent.
• Synthetic options: Preset and designed voices require no real-person cloning sample.
• Existing foundations: Teams can preserve a voice’s core identity while adjusting individual traits.

Consent rules depend on the source voice

Voice replication remains a separate capability that can recreate a voice from a 30-second audio sample. Google AI Studio does not offer replication in Illinois, Texas, the European Economic Area, the United Kingdom, Switzerland, or India. Teams working with a replicated real-person voice still need appropriate consent, while preset and designed synthetic voices avoid the recording workflow and listed geographic restrictions.

Preview endpoints expire in 2026

Google will shut down three older preview models on November 17, 2026:

• Gemini 2.5 Flash Preview TTS
• Gemini 2.5 Pro Preview TTS
• Gemini 3.1 Flash TTS Preview

Google recommends Gemini 3.8 Flash TTS or Flash-Lite TTS as migration targets. Applications using a preview endpoint should test voice selection, prompt behavior, language coverage, latency, and generated audio before changing production traffic.

Try a remix in four steps

1. Open Google AI Studio and enter the TTS interface.
2. Select a preset, designed, or replicated voice.
3. Choose Remix this voice and describe the requested changes.
4. Compare the three takes and keep the preferred result.

Production applications can access the same controls through the Gemini API with gemini-3.8-flash-tts . Supported starting points include prebuilt voices, Extended Voice Library entries, custom voice-design personas, and replicated voices. Google AI Studio also provides interactive workflows for designing and replicating voices.

Teams building voice agents, audiobooks, narration systems, or localized content can use remixing to refine a selected voice instead of repeatedly generating replacements. The workflow is most useful when the base voice already fits the product and only its pacing, tone, accent, or performance needs adjustment.