OpenAI Generate Text

A lot of the interesting ComfyUI workflows these days aren't really about generation - they're about an LLM deciding something, then the graph acting on it. YogurtOpenAIGenerateText is the node that makes that decision: a full OpenAI chat call whose output lands as a STRING you can branch on, feed into a prompt, or log.

The controls you'll actually touch:

• api_key - inline, or blank to fall back to the pack's api_key.json or OPENAI_API_KEY.
• model_name - defaults to gpt-5.1.
• system_prompt and prompt - the system framing and the user message. Multiline, so real prompting fits.
• temperature and top_p - sampling heat and diversity. top_p defaults to 0, which means "not enforced" here.
• max_tokens - output ceiling (default 4096).
• frequency_penalty / presence_penalty - dials for repetition and topic drift, both -2 to 2.
• retry_count, timeout, proxy_url - resilience.
• chat_template - the default template wraps {{system_instruction}} and {{prompt}} in <-system-> / <-user-> markers, and you can restructure it freely.

Optional inputs: history (feed the previous turn's HISTORY output for a multi-turn conversation) and extra (raw JSON added to the request body). Outputs: text (the reply), history (chain it back in for the next call), and payload - the raw response object, for when you need more than the text.

How you'd use it

The classic loop: ask the model to analyze a prompt or an image caption and output a yes/no or a structured answer, then wire text into a switch or conditional. The history output makes real back-and-forth possible - a discussion with the model that spans multiple node calls instead of a single shot. And the base_url flexibility is quietly the best feature: point it at Azure or any OpenAI-compatible endpoint (LocalAI, Ollama, a self-hosted proxy) and this node becomes your universal LLM connector.

Install and keys

Ships in ComfyUI-YogurtNodes, the pack's LLM family (OpenAI, OpenRouter, Gemini, Grok, Qwen, Vertex AI - same client pattern throughout). ComfyUI Manager → search ComfyUI-YogurtNodes → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/yogurt7771/ComfyUI-YogurtNodes.git
cd ComfyUI-YogurtNodes
pip install -r requirements.txt

Restart ComfyUI, look under "Yogurt Nodes/LLM". Requires the openai Python package; no models to download.

Key gotcha: this node does one call per run - it's not a chat session unless you wire history back around. And watch max_tokens vs the model's real output length; a truncation at 4096 tokens can silently chop an answer the downstream node was expecting whole.