Gemini TTS

Every TTS API gives you a voice list. Gemini TTS gives you a voice list and the ability to say "Newscaster," "Whisper," or "Vocal Smile," set the pace to "Rapid Fire," and pick an accent - and it actually follows. This node wraps Gemini's text-to-speech, and it's the pack's most expressive voice option by a mile. If you're making characters, doing voice-overs with a specific read, or narrating with a personality, this is the one to fight with.

The catch is it's a Gemini-preview TTS: quality is good, controls are rich, but it's a "preview" model family, so expect occasional API hiccups and model-name churn as Google shuffles the lineup.

How it works

The node uses the google-genai SDK and calls models.generate_content() configured as a TTS generation - your text, a voice_id from a catalog of prebuilt voices, and a VoiceConfig carrying the style, pace, and accent you chose. The returned audio bytes become a Comfy AUDIO dict (waveform + sample rate). The extra inputs like audio_profile and scene get woven into the generation config for context-aware delivery.

The inputs that matter

• text - what gets spoken.
• voice_id - thirty prebuilt voices with moon- and star-inspired names: Zephyr, Puck, Charon, Kore, Fenrir, Leda, Aoede, Orus, Enceladus, and two dozen more. The names don't tell you the gender or timbre, so you will be auditioning. That's the fun part.
• model - gemini-2.5-flash-preview-tts, gemini-2.5-pro-preview-tts, or the newer gemini-3.1-flash-tts-preview. Flash for speed, pro for polish.
• style - the delivery knob: Vocal Smile, Newscaster, Whisper, Empathetic, Promo/Hype, Deadpan, or None. This is the feature nobody else in the pack has.
• pace - Natural, Rapid Fire, The Drift, Staccato, or None. Pairs with style.
• accent - Neutral, plus American (Gen/Valley/South), British (RP), Transatlantic, and Australian. Yes, you can make a British-RP newscaster.
• temperature (default 1.0) - the expressiveness dial; lower for flat, higher for loose.
• audio_profile and scene (optional) - free-text fields for describing the recording setup/context; leave empty unless you're chasing a specific delivery.
• seed / api_key - key in the field or .env (GEMINI_API_KEY).

The output

audio - one AUDIO output, to preview or save.

Installing it

One of ~20 nodes in the Aryan185/ComfyUI-ExternalAPI-Helpers pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Aryan185/ComfyUI-ExternalAPI-Helpers.git
cd ComfyUI-ExternalAPI-Helpers
pip install -r requirements.txt

Restart ComfyUI; find "Gemini TTS" under audio/generation. Needs google-genai. Gemini key from aistudio.google.com - paste in the field or set GEMINI_API_KEY in .env.

Where people get burned

Preview-model churn is the main headache: Google renames these TTS models, and the enum in the node can point at a name that's been deprecated or quota-limited server-side. If you start getting model-not-found errors, check whether a newer TTS model name is available. Also, the voice catalog is big but opaque - no way to preview in-node, so audition by generating one line per candidate. And TTS via API is billed; the preview tier has limits, so a long narration run can stall on quota.

For "I need a character voice with attitude," this beats the pack's other TTS options. For "I need a reliable studio narration," ElevenLabs or plain OpenAI TTS are the boring, dependable choices.