Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- Voice remixing creates a new voice_... resource without overwriting the source; deleting the base voice does not delete voices derived from it.
- Use POST /v1beta/voices or client.voices.create with store=true, type=prompted, prompted.base_voice and a relative prompted.input.
- The new ID works for single-speaker synthesis with gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts. Audition its 24 kHz WAV preview and change only one or two traits per iteration.
- Remixing requires a Paid Tier 1 or higher Gemini API project. Each project can hold up to 200 shared stateful voices, which expire after a year without use.
- Standard TTS input and audio-output prices are not a confirmed remix creation or storage fee. Check regional eligibility, age and professional-use rules, and the data terms separately.
Remixing creates a new voice instead of rewriting the base
Gemini API released voice remixing for Gemini 3.8 TTS on November 4, 2026. Choose an existing voice and describe relative changes in natural language, such as pitch, perceived age, timbre, energy or accent. The system creates and stores a new voice_... resource.
The new voice is independent of its base: the original resource is not overwritten or changed, and deleting the base later does not delete voices derived from it. This workflow designs a supported synthetic voice; it is not the same as copying a real person's voice.
Create and audition a new voice ID
Create a voice with POST /v1beta/voices or client.voices.create. Set store to true and voice.type to prompted, then put base_voice and input inside voice.prompted. You can set voice.model explicitly to gemini-3.8-flash-tts; it is also the default when omitted. base_voice identifies an existing voice by name or ID, while input describes the relative changes you want.
The response includes the new stored voice_... ID and a sample_audio preview in 24 kHz WAV format. Play the preview before adding that ID to your workflow; the prompt text cannot replace listening to the result.
Choose a supported base and refine it in small steps
A base can be any of the 30 prebuilt studio voices, such as Puck; a catalog voice ID returned by the Extended Voice Library; or a prompted voice_... ID from Voice Design or an earlier remix that belongs to your project. The latter two types can serve as the base for further, step-by-step refinement.
Change only one or two traits per iteration, then audition the preview before deciding whether to continue. For a hypothetical synthetic example, start with the prebuilt Puck voice and ask for a calmer delivery and slightly lower pitch. This is an adjustment to a prebuilt synthetic voice, not a clone of a real person. Use voice remixing for lasting changes to the baseline, such as pitch or accent; use speech_metadata.style for temporary, turn-level performance directions.
Reuse, inspect and clean up stored voices
For single-speaker speech synthesis, put the voice_... ID in speech_config and use either gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts. Listing and retrieving voices lets you inspect a remixed voice's base and prompt; retrieving an individual voice also returns its WAV preview for another listen.
A project can store up to 200 stateful voices, with Voice Design, remixed and stored replicated voices sharing the quota. A voice expires one year after its last use. Synthesizing with it or referencing it as base_voice in a new remix restarts that one-year window; check expire_time when planning cleanup. Before deleting, verify the full voice ID and its dependencies, then remove only the specific resource you have confirmed is no longer needed.
Check project and regional eligibility first
Voice remixing requires a Paid Tier 1 or higher Gemini API project. A consumer subscription is not the same as this API project requirement. The service is for developers aged 18 or older building for professional or business purposes, not consumer use. An API client website, application or other service must not be directed to, or likely to be accessed by, people under 18.
The Gemini API and Google AI Studio are available only in the countries and territories on Google's current region list. API clients serving users in the European Economic Area, Switzerland or the United Kingdom may use Paid Services only. A listed region is not a promise that remixing has been enabled for every account or project. Check the project's eligibility and actual availability; changing a network route does not establish project eligibility.
Separate TTS synthesis rates from remix fees
The published Standard-tier TTS rates are charged per million tokens for input and audio output. Through December 31, 2026, gemini-3.8-flash-tts is USD 0.50 for text input and USD 9 for audio output; gemini-3.8-flash-lite-tts is USD 0.50 for text input and USD 6 for audio output. Starting January 1, 2027, input is USD 1 for both models, with audio output at USD 18 for Flash and USD 12 for Flash-Lite.
These are speech-synthesis model rates, not confirmation of a one-off voice-remix creation or storage fee. The cited material does not say whether such a fee is charged, so treat it as unknown; a free-tier TTS price row does not prove that voice remixing is free. Check the rate tier that applies to your project and the latest pricing before estimating a bill.
Review the terms and data handling before use
Use of the Gemini API remains subject to applicable law, the Google APIs Terms, the Gemini API Additional Terms and the Prohibited Use Policy. Paid Services do not use prompts and responses to improve products, but the terms still allow them to be logged for a limited period to detect and prevent policy violations, maintain service safety and security, and make required legal or regulatory disclosures. Data may also be stored transiently or cached in countries where Google or its agents operate facilities. Paid access does not mean zero retention.
Before production use, check current project requirements, pricing and terms against the text, voice use case and users’ locations, then assess the results with representative material of your own.
Sources
Frequently Asked Questions
Will the base voice's accent change automatically?
The base voice's language and regional accent are generally preserved. Ask for a different language or accent explicitly in prompted.input when you want to change them.
Does retrieving metadata or listening to a preview extend the expiry?
The documented renewal triggers are synthesizing speech with the voice or using it as base_voice in a new remix. The guide does not list a metadata read or preview playback alone as a renewal trigger. Check expire_time instead of assuming that opening a preview extended it.