> For the complete documentation index, see [llms.txt](https://docs.bottalk.io/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.bottalk.io/docs/tts-providers/gemini.md).

# Gemini

## Gemini Vertex Endpoint

The Gemini Vertex endpoint runs Google's Gemini text-to-speech models on Vertex AI. It is selected per project from **Project Settings → Voice Selection**.

### Enabling the endpoint

1. Open the project you want to switch.
2. Go to **Project Settings → Voice Selection**.
3. In the **Voice** dropdown, choose a voice with the `-GMVA` suffix (for example, `de-DE-Charon-GMVA`). Selecting one of these voices activates the Gemini Vertex endpoint for the project.
4. Under **Model**, pick one of the Gemini TTS models (see below).
5. Under **Generation parameter**, choose **Seed** or **Temperature**. The endpoint uses one parameter at a time, and **Seed** is selected by default.
6. Save.

New articles published in this project narrate on the new endpoint. Audio rendered before the change keeps its previous configuration.

### Models

<figure><img src="/files/f8XvN6i9ONdyl6oysR20" alt=""><figcaption></figcaption></figure>

Three models are available in the **Model** dropdown:

* **`gemini-2.5-pro-tts`** — highest fidelity, slowest. Suitable for long-form pieces where extra nuance is worth the additional latency.
* **`gemini-2.5-flash-tts`** — default. Balanced quality and speed. Recommended for most projects.
* **`gemini-3.1-flash-tts-preview`** — preview model. Behaviour can change without notice while it is in preview. Use it only if you intend to track preview updates.

### Generation parameters

The endpoint accepts **one** generation parameter at a time: **Seed** or **Temperature**. Switch between them with the toggle under **Generation parameter**.

<figure><img src="/files/ocBZYguHDJVuvJW5Yug3" alt=""><figcaption></figcaption></figure>

#### Seed (default)

Seed is a number per project that anchors the model's output. With the same article, model, and seed, the endpoint returns the same audio on every render — including re-renders later in the project's life.

Each project has its own seed. The current value is shown in the input below the toggle, and can be replaced with any integer.

Keep Seed when consistency matters: when a returning listener should hear the same delivery on a re-render, and when new articles should sit alongside earlier ones without a shift in voice or cadence.

#### Temperature

Temperature lets the model vary its delivery between renders. The same article can produce noticeably different audio on each render, with variation in cadence and intonation.

Switch to Temperature when variation is the goal — for example, when sampling alternative deliveries during voice selection, or in creative formats where uniform output is not wanted.

Re-rendering an article on Temperature does **not** reproduce previous audio.

### Switching the parameter

Toggling between Seed and Temperature applies to **future renders only**. Audio already rendered keeps the parameter it was generated with. To bring older articles onto the new parameter, regenerate them explicitly.