✨ OpenAI Configurator

The OpenAI Configurator is the Gemini configurator's twin, and the first node you drop when you want the OpenAI side of this suite. It holds your API key, picks the model, sets the voice and reasoning effort, and packs it all into one openai_model connection for the 🚀 OpenAI API Runner. Nothing executes on its own - it's pure configuration - but every OpenAI run in the pack flows through it.

The inputs that matter

• api_key - your OpenAI platform key. Rendered as password dots by the pack's frontend so it doesn't leak during screen recordings, and the node hard-stops with a ValueError if you run it empty. There's also a "🔄 Reload Model" button the JS adds, so you can swap keys without restarting ComfyUI.
• model_name - eight choices, default gpt-5.4-mini. That default is a deliberate pick: cost-effective and fast, the pack's own recommendation for most in-graph work. gpt-5.4-nano is even cheaper; gpt-5.4 and gpt-4o are the quality step up; o1-mini is the reasoning model - remember from the runner article that the o1 family folds your system prompt into the user message and refuses images.
• generate_audio - the boolean that decides whether TTS happens at all. Off by default. If you leave it off, the runner still returns a response_audio output, but it'll be a silent empty dict - the classic "why is there no sound?" trap. Flip this on only when you actually want a voice.
• voice_name - the six OpenAI TTS voices: alloy, echo, fable, onyx, nova, shimmer (default alloy). These are the standard tts-1 personas - cleaner and more neutral than the Gemini Gen-Audio set, which is a fair summary of the two packs' voice personalities. Only matters when generate_audio is on.
• reasoning_effort - default, low, medium, high. Passed through to the API when it's not default; dial it up for harder questions, down for speed and cost.

Output

One output: openai_model (OPENAI_MODEL) - the dict carrying the client, model name, key, TTS flag, voice, and reasoning effort into the runner. That's the only connection you make.

Install

Pack-standard: ComfyUI Manager → search "Universal LLM Suite" → Install, or:

cd ComfyUI/custom_nodes/
git clone https://github.com/alice-ai-wonder/ComfyUI-Universal-LLM-Suite

Restart and it auto-installs openai, google-genai, and librosa on first boot. No model downloads - this is API-only, so everything runs at OpenAI's end and you pay per token. You'll need a key from platform.openai.com with a funded account.

Common issues

• "API key must not be empty" - the one hard failure; paste a real key.
• Silent response_audio - generate_audio is still off. It's easy to miss because the toggle sits in the configurator, not the runner.
• o1 surprise - if you picked o1-mini, vision input and a true system message both go away; switch to gpt-4o / gpt-5.4 for those.

The pack is young and barely known (one commit, essentially no community discussion as of mid-2026), so it carries the standard custom-node caveat: arbitrary Python runs on load, auto-installing packages. Give the source a skim before you paste a paid API key into it.