ComfyUI OpenRouter

This is a ComfyUI extension that runs any model on OpenRouter with your own API key. It has six kinds of nodes:

• Chat: ask a model a question, with images, video, audio, or documents. Get back text, images, or speech.
• Image: generate and edit images, including masks and SVG files.
• Video: make a video from a prompt, frames, or reference media.
• Audio: turn text into speech, optionally in a cloned voice, and speech into text with timestamps and subtitles.
• Search: turn text and images into embeddings, and rank them against a query.
• Decisions: ask Jev, OpenRouter's decision model, questions it answers with probabilities. A workflow can branch on the answers.

Requests go from your ComfyUI server to OpenRouter, which passes them to the model's provider and bills your OpenRouter account. Every run of a paid node is billed, so check the model's price in OpenRouter models before you run it.

Contents

• Requirements
• Install
• First run
• Example workflows
• Nodes
• Models
• How a run works
• Development
• Common problems
• License

Requirements

• Python 3.12 or later.
• ComfyUI 0.34.6 or later, with frontend 1.49.6 or later in the 1.x series.
• An OpenRouter account with credit, and an API key from openrouter.ai/settings/keys.
• Network access from the ComfyUI server to openrouter.ai.

Install

1. Stop ComfyUI.
2. Put the extension in ComfyUI/custom_nodes/comfyui-openrouter, with __init__.py and requirements.txt directly inside that folder.
3. Install requirements.txt with the Python that runs ComfyUI, using the command for your installation below.
4. Start ComfyUI and reload its window.

Comfy Desktop

1. On the home screen, open the installation's ⋮ menu and select Manage.
2. Under About, copy Location. In that folder, find the ComfyUI folder that holds main.py, and put the extension in its custom_nodes folder.
3. Select Terminal in the same panel. It opens in the ComfyUI folder with the installation's Python. Run:

pip install -r custom_nodes/comfyui-openrouter/requirements.txt

Then start the installation. Comfy Desktop's Manage panel explains these controls.

macOS or Linux

From the ComfyUI folder, replacing .venv if your environment has another name:

.venv/bin/python -m pip install -r custom_nodes/comfyui-openrouter/requirements.txt

With uv:

uv pip install --python .venv/bin/python -r custom_nodes/comfyui-openrouter/requirements.txt

Windows

From the ComfyUI folder, in PowerShell:

.\.venv\Scripts\python.exe -m pip install -r .\custom_nodes\comfyui-openrouter\requirements.txt

Windows portable

From the folder that holds ComfyUI and python_embeded, in PowerShell:

.\python_embeded\python.exe -m pip install -r .\ComfyUI\custom_nodes\comfyui-openrouter\requirements.txt

To update or remove the extension, see the advanced guide.

First run

1. Open ComfyUI menu → Extensions → OpenRouter → OpenRouter settings and save your API key. It is stored on the server, in the extension's private folder.
2. Open Browse Templates → comfyui-openrouter → decision-01-verify-then-escalate, or drag its file onto the canvas.
3. Select Run.
4. Read the answer in Answer, and Jev's check in Summary.

To stop a run, use ComfyUI's cancel button. To send the same request again, change run number. Each request's charge is listed at openrouter.ai/activity.

Example workflows

There are 13 example workflows, and together they use every node. Most of them have Jev decide something: which image to keep, whether a draft is distorted, or whether an answer is backed by its sources.

Open one from Browse Templates → comfyui-openrouter, or drag its file onto ComfyUI. Each group has a note that says what it does; start with Start Here. Each step after the inputs is a subgraph: select the icon at its top-right corner to open it. Media inputs start empty, so add your own image, audio, or document.

Some workflows use ComfyUI's If/Else Switch (beta). It runs only the branch it picks, so you pay only for the paid nodes on that branch.

After an update, open the examples in a new tab to get the new versions.

Chat

Images

Video

Audio

Search

Decisions

Request Options is used in Chat: Write a Product Listing, to keep the listing away from providers that store data.

Nodes

The nodes are in the OpenRouter menu, under Chat, Image, Video, Audio, Search, and Decision. ComfyUI's node help opens each node's guide.

The decision nodes use OpenRouter's alpha decisions API (/api/alpha/decisions), which OpenRouter may still change.

Models

The extension ships with a copy of OpenRouter's model list. To get the current list and prices, open Extensions → OpenRouter → OpenRouter models and select Refresh Models. The node dropdowns update after a refresh. The same dialog estimates prices.

To use a model that is not in the list, choose other model ID in a node's model dropdown and type its ID. You can add a suffix, such as :online for web search.

Automatic checks tell you when OpenRouter's list has changed; select Refresh Models to take the new list. See model updates.

How a run works

Node inputs ──▶ ComfyUI server (the key and settings stay here)
                  │ checks the inputs against the chosen model
                  ▼
               OpenRouter ──▶ the model's provider
                  │
                  ▼
Node outputs ◀── text, images, video, audio, and answers

1. The node checks its inputs against the chosen model's capabilities and stops with an error on an input the model cannot take. This happens before anything is sent or billed.
2. It waits for a free slot. At most parallel requests requests run at once; the default is 4.
3. It sends one request from your ComfyUI server to OpenRouter with your key. OpenRouter passes it to the model's provider and bills your account.
4. For video, the node records the job, checks it until the video is ready, and downloads it. The checks and the download are free.
5. The node turns the reply into ComfyUI outputs. Nodes connected to an empty output, such as images from a text-only model, are skipped.
6. Your save nodes write the files. ComfyUI reuses a result while the inputs are unchanged; change run number to send the request again.

The advanced guide lists what each node sends and returns.

Development

Development needs Git and mise. Tell the checks where ComfyUI is, in .mise.local.toml:

[env]
COMFYUI_PATH = "/path/to/ComfyUI"
COMFYUI_PYTHON = "/path/to/ComfyUI/.venv/bin/python"

Install the tools, dependencies, and Git hooks, and run the full check:

mise trust
mise run repo:setup
mise run repo:check

Restart ComfyUI after changing Python. After changing the browser code, run mise run comfy:frontend:build; after changing a workflow description, run mise run comfy:workflows:build. The advanced guide lists every task.

Common problems

For the messages a node shows, see troubleshooting. The advanced guide also covers keys, settings, limits, prices, model updates, caching, and video recovery.

License

The code, guides, and workflows are under the MIT license. Third-party components keep their own licenses.