Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.comfy.org/llms.txt

Use this file to discover all available pages before exploring further.

OpenRouter is a unified routing platform for large language models. It exposes many providers through one API so you can switch models without changing your integration. ComfyUI connects through the OpenRouter LLM partner node, which offers a curated list of text and multimodal models—Claude, GPT, Gemini, Grok, DeepSeek, Qwen, Mistral, GLM, Kimi, and Perplexity Sonar—in a single node.
To use the API nodes, you need to ensure that you are logged in properly and using a permitted network environment. Please refer to the API Nodes Overview section of the documentation to understand the specific requirements for using the API nodes.
Make sure your ComfyUI is updated.Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated. (Desktop version’s update will delay sometime)If nodes are missing when loading a workflow, possible reasons:
  1. You are not using the latest ComfyUI version (Nightly version)
  2. Some nodes failed to import at startup

​
What you can do

  • Text chat — prompts and optional system_prompt for role or task instructions
  • Vision — attach reference images on supported models (slot count depends on the model)
  • Video input — Gemini 3.5 Flash and Qwen 3.6 Plus/Flash support optional reference videos
  • Reasoning — reasoning_effort on reasoning / frontier-reasoning models (off, low, medium, high)
  • Web search — Perplexity Sonar models expose search_context_size for grounded answers
Token pricing is listed under Partner Node Pricing.

​
Supported models

ModelVision (max images)VideoExtra options
anthropic/claude-opus-4.720—Reasoning effort
openai/gpt-5.5-pro20—Reasoning effort
openai/gpt-5.520—Reasoning effort
google/gemini-3.5-flash20Up to 4Reasoning effort
x-ai/grok-4.2020—Reasoning effort
x-ai/grok-4.320—Reasoning effort
deepseek/deepseek-v4-pro——Reasoning effort
deepseek/deepseek-v4-flash——Reasoning effort
deepseek/deepseek-v3.2——Reasoning effort
qwen/qwen3.6-max-preview——Reasoning effort
qwen/qwen3.6-plus10Up to 4Reasoning effort
qwen/qwen3.6-flash10Up to 4Reasoning effort
mistralai/mistral-large-25128——
mistralai/mistral-medium-3-58—Reasoning effort
z-ai/glm-4.6——Reasoning effort
z-ai/glm-5——Reasoning effort
moonshotai/kimi-k2.610—Reasoning effort
moonshotai/kimi-k2-thinking——Reasoning effort
perplexity/sonar-pro——Search context size
perplexity/sonar-reasoning-pro——Search context size, reasoning effort
perplexity/sonar-deep-research——Search context size, reasoning effort

​
Model-specific inputs

The model dropdown is a dynamic combo. When you change the model, the node updates which inputs appear:
  • Image slots — shown only for vision-capable models; the number of slots matches the model limit
  • Video slots — shown only on Gemini 3.5 Flash and Qwen 3.6 Plus/Flash
  • Reasoning effort — shown on reasoning and frontier-reasoning models
  • Search context size — shown on Perplexity Sonar models
Text-only models hide image and video inputs.

​
Example workflow (api_openrouter_llm)

The template below is one example, not the only way to use the node. Download it to see how inputs connect, then adapt the graph for your task—pure text chat, captioning, prompt rewriting, feeding strings into image nodes, or anything else that fits your pipeline.

Download Workflow

Download JSON or search “OpenRouter LLM” in Template Library
In this sample, the workflow analyzes a reference image and returns a text-to-image prompt for downstream image nodes. You can omit Load Image, change the prompts, or rebuild the graph entirely for your own use case.
If you use the template as a reference, a typical run looks like this:
  1. In Load Image, load the reference image (skip for text-only chat)
  2. In OpenRouter LLM, connect the image to image_1 (more slots appear when the model supports them)
  3. Set prompt — e.g. ask the model to analyze the image and output a generation prompt
  4. (Optional) Set system_prompt in advanced options for tone, format, or constraints
  5. Pick a model; adjust reasoning_effort or search_context_size when shown
  6. Click Run or press Ctrl(cmd) + Enter
  7. View the reply in Preview Any

​
Additional notes

  • seed — set to 0 to omit; most models treat it as a hint only
  • Reference images and videos are uploaded automatically when the workflow runs; you only wire ComfyUI image/video outputs into the node
  • For pricing per model, see OpenRouter on the pricing page