Gemini 3.1 Flash TTS
Google's text-to-speech for voiceover and narration — 30 prebuilt voices, single- or two-speaker dialogue, and inline audio tags that steer delivery style. Up to 3 minutes per run, ~10 second turnaround.
Capabilities
| Feature | Support |
|---|---|
| Text-to-Speech | Yes |
| Voices | 30 prebuilt |
| Speakers | 1 or 2 per run |
| Style Control | Inline audio tags |
| Max Duration | 180 seconds |
Audio Tags
Drop bracketed delivery cues anywhere in the script:
[excited] We actually did it! [whispers] Don't tell anyone yet.
[shouting] Everyone — dinner's ready!
Tags change the read from that point on: [excited], [whispers], [shouting], [sighs], [laughing] and similar natural-language cues all work.
Two-Speaker Dialogue
Assign a voice to each of two named speakers and write a script with speaker labels — the model renders the conversation in one continuous, naturally-timed track. Ideal for interview-style narration, ad dialogues, and animatic scratch tracks.
Prompting Tips
- Audition voices with a short line first. Voice choice dominates the result; lock it before rendering the full script.
- Punctuation is pacing. Commas, em-dashes, and ellipses read as pauses; short paragraphs create natural breath points.
- Combine tags with context. "[tired] It's been a long day…" — the tag sets delivery, the text keeps it natural.
Limitations
- Preset voices only — no reference-audio input
- Max 2 speakers per run
- 3-minute ceiling per generation
See Also
- Qwen3 TTS — 36 voices including 8 Chinese regional dialects
- Seed Audio 1.0 — reference-audio-guided multimodal audio