☁️RunningHub Workflow

RHWorkflow (display name "☁️RunningHub Workflow") is the "rent someone else's GPU" move, and it's the most powerful node in ComfyPanel. Where RHWebApp runs pre-packaged cloud Apps, this node takes a workflow - your workflow, or one you downloaded - and executes it on RunningHub's cloud, then pulls the results back into your local graph as files on disk. You keep your ComfyUI frontend; RunningHub just supplies the compute and the 24G/48G GPUs you don't own.

This is the exact case the ComfyUI ecosystem knows well from the cloud-compute side of the house: the workflow is fine, your hardware isn't. RunningHub is a genuinely established platform - hundreds of community threads treat it as a ComfyUI-native cloud with a full native API - and RHWorkflow speaks to it through that native API rather than a screen-scraping shim.

How it works

One required input, workflow_file (a STRING, default empty), and the pack's description spells out the two ways to use it:

• Mode A - online: paste a 19-digit RunningHub Workflow ID into workflow_file. The node fetches the workflow JSON from RunningHub's API, converts the UI graph into API format, and runs it in the cloud.
• Mode B - local file: give it a path to a workflow .json you downloaded from RunningHub, and it loads and runs that.

The 19-digit check is how it tells the modes apart - digits of exactly that length mean online, anything else is treated as a file path. From there it's a standard async pipeline: override any parameters, upload any image/video inputs, submit to RunningHub's ComfyUI proxy (/proxy/<key>/prompt), poll /history every five seconds until the job finishes, download the output images/gifs/videos/audio, and emit them across 30 * outputs, ordered by the producing node's ID. Progress reports land on the node while it waits, and it can be interrupted like a local run.

Parameters worth knowing

Beyond workflow_file, the node lets you override values inside the remote workflow through dynamic inputs named param_<nodeid>_<fieldname> - param_5_seed would set the seed widget on node 5, for instance. And anything you connect that ends in _<nodeid> (like an image destined for a specific node) gets uploaded to RunningHub and injected into that node's input. It's a clever scheme, but you'll mostly discover those by poking at the generated sockets on the node.

What you need

Same credentials as RHWebApp: right-click → ⚙️ RunningHub Settings, enter your API Key, Shared API Key, and User ID. That config is shared pack-wide. Also remember this is metered cloud compute - you're paying in RunningHub credits per run, and polling can go up to roughly half an hour (360 attempts × 5s) before it times out, so budget accordingly for slow jobs.

Common issues

An empty response with "no prompt_id returned" is RunningHub's classic failure combo: invalid/expired key, insufficient credits, or a malformed workflow JSON - the node's error text lists exactly those four causes. If workflow_file isn't a 19-digit number, it will try to read it as a local file path and throw FileNotFoundError when it's missing - that's the most common beginner stumble. And in local mode, the file must live on the machine running ComfyUI, not on your desktop elsewhere.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/Ginolazy/ComfyPanel

Restart ComfyUI, or search ComfyPanel in ComfyUI Manager. Requirements are the usual torch/numpy/Pillow plus requests, aiohttp, kornia, scipy, opencv-python. No local model downloads - compute happens remotely. The Photoshop plugin is optional.