OpenAI Text-to-Speech

OpenAI's TTS API is the middle child of this pack's voice options: cheaper and faster than ElevenLabs, more conventional than Groq Orpheus, and the one with the newest trick - gpt-4o-mini-tts, which takes written instructions about how to speak, not just what to say. This node wraps all three TTS models behind one ComfyUI node and hands you back a ready AUDIO output.

If you just need a clean narration track and don't need to nerd out over voice identity, this is a perfectly good default. The real reason to pick it over the cheaper Groq option is the instruction-following on gpt-4o-mini-tts: "speak slowly, like you're explaining to a child, slightly excited" actually changes the delivery.

How it works

The node calls OpenAI's audio.speech.create() with your text, model, voice, format, and speed, and decodes the returned audio bytes into a Comfy AUDIO dict (waveform + sample rate). One request, one audio clip back. The instructions field is only sent when the model supports it - the node checks and refuses to send it to tts-1/tts-1-hd, which would error server-side.

The inputs that matter

• text - what gets spoken.
• model - gpt-4o-mini-tts (the steerable one, and the modern default), tts-1, or tts-1-hd (the older pair; hd is higher quality, both plain).
• voice - nine voices: alloy, ash, coral, echo, fable, onyx, nova, sage, shimmer. They're preset personas, not cloneable. Alloy and nova are the usual favorites; the newer ash/coral are worth a listen.
• response_format - mp3, opus, aac, flac, wav, or pcm. WAV/PCM for lossless editing, mp3 for small files, opus for the best quality-per-byte.
• speed - 0.25 to 4.0, default 1.0. Wide range; 0.5–1.25 is the useful band.
• instructions (optional) - only works with gpt-4o-mini-tts. Free-text direction on delivery: tone, pacing, emphasis. This is the feature that makes the node worth using.
• api_key - OpenAI key in the field or .env (OPENAI_API_KEY).

The output

audio - one AUDIO output. Preview it or save it via a save-audio node (Video Helper Suite has you covered).

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 "OpenAI Text-to-Speech" under audio/generation. The openai dependency is the one that matters. You need an OpenAI API key with billing; paste it in or set OPENAI_API_KEY in .env (copy .env.example).

Where people get burned

The instructions field is the trap: fill it in while running tts-1 or tts-1-hd and the node raises an error on purpose (the API doesn't support instructions there). Swap to gpt-4o-mini-tts or leave it empty. Also, these are preset voices only - there's no voice cloning, so if you need your voice or a specific one, ElevenLabs is the node for that. And it's a paid API: TTS is cheap per character but a long script still bills, so do your iteration at lower speed and mp3 before the final high-quality pass.

For a no-fuss narration node, this is the one I'd wire up first. For expressive or cloned voices, the pack's other TTS nodes have their own strengths.