RH Execute

This is the one that does the work. Everything else in the ComfyUI_RH_APICall pack - the Settings node, the uploaders, the NodeInfoList overrides - exists to feed this node. RH_ExecuteNode is the trigger: it fires a workflow (or an AI WebApp) on RunningHub's servers, waits while it runs on their GPU, streams a progress bar into your ComfyUI, and drops the finished results back onto your local canvas as ordinary types. If you only wire up one RunningHub node knowingly, it's this one.

The pack as a whole lets your local ComfyUI run jobs on RunningHub, a cloud ComfyUI host, instead of your own hardware. This node is where "run it in the cloud" actually happens. It's an output node, so it's the endpoint of the graph - you don't wire it into anything downstream, you read your results off it.

How it works

You hand it your credentials and a set of overrides, it calls the RunningHub API to start the task, and then it polls until the job finishes (a recent update added an automatic fallback from WebSocket to HTTP polling so a flaky connection doesn't kill the whole run mid-job). When the remote workflow completes, whatever it produced comes back down the wire and gets sorted into the matching output socket. Because it's a real network round-trip to a rented GPU, a run takes as long as the cloud job takes - this isn't instant, and it isn't local.

The inputs that matter

• apiConfig (STRUCT, required) - the bundle from RH_SettingsNode: your key, base URL, and the workflow/webapp ID. This is the only required input; without it there's nothing to run.
• nodeInfoList (ARRAY, optional) - your overrides, built with the pack's NodeInfoList node. This is how you drive the cloud workflow from your machine: swap the prompt, bump the seed, point a load-image node at a file you uploaded. Leave it empty and the remote workflow runs with whatever defaults it was saved with.
• is_webapp_task (boolean, default false) - flip this on if the ID in your Settings node is an AI WebApp rather than a plain workflow. Get it wrong and the task won't start.
• run_timeout (default 600 seconds) - how long to wait before giving up on the remote job. Bump it for slow video or large batches.
• concurrency_limit (default 1) - how many tasks to run at once.
• use_rtx4090_48g (boolean, default false) - request the 48GB RTX 4090 tier on RunningHub's side, for jobs that need the extra VRAM.

The outputs

This node has ten output sockets so it can carry back whatever the cloud workflow made: images (IMAGE), video_frames (IMAGE), latent (LATENT), text (STRING), audio (AUDIO), and video1 through video5 (five separate VIDEO ports, added so a workflow that returns multiple videos can hand them all back at once). You wire up only the ones your workflow actually produces - a text-to-image job uses images and ignores the rest.

Installing it

• ComfyUI Manager: search ComfyUI_RH_APICall, install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/HM-RunningHub/ComfyUI_RH_APICall, then restart ComfyUI.

No local models or heavy deps - the GPU work is remote. Set your key and workflow ID on the RH_SettingsNode, and start from a graph in the pack's examples/ folder so you can see the Settings → uploaders → NodeInfoList → Execute chain already assembled.

Where people get burned

If it errors instantly, it's credentials - check the Settings node, especially the .cn/.ai domain trap. If the task starts but returns nothing useful, your overrides missed: a nodeId or field name in the NodeInfoList that doesn't match the cloud workflow, so it ran on defaults. If it dies partway, raise run_timeout - video and big batches routinely blow past 600 seconds. And remember every run is billed cloud compute on RunningHub and ships your prompts and inputs to their servers, so it's a genuinely different trust and cost model than running the same graph locally.