Nodes/ComfyPanel/☁️RunningHub App
ComfyUI Node

☁️RunningHub App

Run a cloud App like Sora or GPT Image from inside your graph

By Ginolazy·Created 6 months ago·Updated 2 days ago· 23
☁️RunningHub App
    • Result
    APP

    This node calls a cloud service instead of running a model - the "the model isn't on your machine" move, done cleanly. RHWebApp (display name "☁️RunningHub App") is a dropdown that lists pre-packaged AI apps hosted on RunningHub, a cloud GPU platform. You pick an app - think Grok Image, GPT Image, Sora-class video, the sort of closed models you literally cannot download - set its parameters in automatically-generated widgets, hit run, and the result lands back in your graph as if a local node had made it.

    It's the exact pattern the ComfyUI ecosystem knows as an API-wrapper node, and it's the sensible pick when your hardware can't run the job or the model has no open weights at all. RunningHub is a real, established platform - it shows up across hundreds of community threads as a ComfyUI-native cloud with full API access - and this node is its front door from inside ComfyUI.

    How it works

    The required APP input is a dropdown, but it isn't a fixed list. The node's frontend fetches RunningHub's app catalog - a default list of apps plus anything on your account - and builds the dropdown from that. Select an app and the node queries its definition and generates parameter widgets on the fly: sliders, toggles, text fields, and image/video upload fields where the app takes media. That's why a 0-input-schema node feels like it has a hundred inputs; they materialize based on the app.

    Under the hood, execution is a straightforward pipeline:

    1. Upload any image/video inputs to RunningHub (it handles tensors, numpy arrays, and files).
    2. POST the task to RunningHub's task/openapi/ai-app/run endpoint.
    3. Poll task/openapi/outputs every few seconds until the job completes.
    4. Return the result URLs as a list through the single Result output (marked as a list).

    Progress shows on the node, and it can be cancelled mid-poll like a local run.

    What you need to use it

    An API key. Right-click the node → ⚙️ RunningHub Settings, and enter your RunningHub credentials - API Key plus Shared API Key and User ID, which you get from the RunningHub site's API Keys tab. The pack reads that config for both this node and RHWorkflow. There's also geo-routing baked in: it targets runninghub.cn or runninghub.ai depending on your account region.

    Two things to know before you dive in. First, this is metered - every call spends RunningHub credits, and video apps can burn through them fast, same as any hosted model. Second, it's a network+credential node, which is exactly the shape of thing worth double-checking you installed from the official repo - git clone https://github.com/Ginolazy/ComfyPanel from the real source, not a fork.

    Common issues

    Empty dropdown → hit refresh on the app list, and make sure your API config is saved. "No App selected" / "Missing web_app_id" → reload the node so it can re-fetch the app definition. A failed task usually means one of three things: an invalid or expired key, not enough credits, or you're pointing at the wrong region endpoint. The node will generally surface the platform's own error message, which is more useful than it sounds.

    Installing it

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

    Restart ComfyUI, or search ComfyPanel in ComfyUI Manager. The pack's requirements (torch/numpy/Pillow, requests, aiohttp, kornia, scipy, opencv-python) are largely already present; Manager will prompt for what's missing. No model downloads - that's the whole point. The Photoshop plugin is optional; this node works standalone.

    CategoryComfyPanel

    Inputs (1)

    NameTypeDefaultDescription
    APPCOMBO0 options:

    Outputs (1)

    NameTypeDescription
    Result*