OpenRouter Generate Text

Every serious LLM workflow eventually runs into the same question: which model am I actually paying for, and is it the right one for this job? YogurtOpenRouterGenerateText is the node that answers "any of them" - it calls OpenRouter's text API, so the model is just a slug you type, and swapping from Claude to GPT to some obscure open-weights runner is a one-string edit.

The controls, in the order you'll set them:

• api_key - OpenRouter key, or blank to fall back to api_key.json / OPENROUTER_API_KEY.
• model_name - defaults to anthropic/claude-3.5-sonnet, but any OpenRouter text model slug works.
• system_prompt and prompt - the usual framing + user message, both multiline.
• temperature, top_p, max_tokens (default 8192) - sampling dials.
• infrastructure_provider - "auto" or a specific one (azure, aws, ...).
• provider_list - comma-separated providers, e.g. openai,azure,together; empty means use infrastructure_provider.
• seed, retry_count, timeout, proxy_url - determinism and resilience.
• chat_template - the <-system->/<-user-> template, restructurable.

Optional: history for multi-turn and extra for raw JSON body additions. Outputs: text, history, payload.

How to think about it

The payload output is worth calling out - it's the raw response object, which means you can dig out metadata (token usage, model actually served, finish reason) that the plain text output hides. For a workflow that needs to log costs or verify which model answered, that's gold.

The model-swapping story is the real pitch. A/B testing two models on the same prompt is changing a string and re-running. When OpenRouter's routing picks a fallback provider, you barely notice. The flip side of that flexibility: the routed model may not be the one you typed - OpenRouter can substitute within its rules - and payload is how you confirm what actually ran.

Install and keys

Ships in ComfyUI-YogurtNodes. 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, find it under "Yogurt Nodes/LLM". Needs the openai package and an OpenRouter key.

Two practical notes. First, the default claude-3.5-sonnet is an older, cheaper model - fine for plumbing, but check OpenRouter's current catalog for what's best today. Second, max_tokens at 8192 is generous; if you're building a loop that calls this node per image, per-item cost adds up fast. Keep the ceiling tight and you'll thank yourself later.