RemRun Set Num Outputs

Every node in this pack exposes ten input and ten output sockets by default. That's a deliberate number - enough for most graphs, and it keeps the nodes from looking like a wall of pins. But ten isn't always enough. If you're feeding a remote run with fifteen conditioning inputs, or pulling a dozen results back out of the far side, you need more slots. This node is the lever: a single integer that changes how many inputs and outputs all the Remote Run nodes expose.

It's a one-trick node, and it knows it. One required input, num, default 10, min 1, max 100. No outputs. It's marked as an output node so running it registers as a "done" step. You set the number, run the node, and the pack's global socket count changes for the session.

How it works

Under the hood, every Remote Run node sizes its input/output lists from a single module-level value at definition time. This node just overwrites that global with your num and logs the change:

Changing RemoteRun number of inputs/outputs from 10 to 20

Because it changes a global, it affects every Remote Run node in the pack at once - that's the design, not a bug. You can't give one node 20 sockets and leave the rest at 10.

The one thing to know

The frontend caches node definitions, so changing num at runtime may not immediately repaint the sockets of nodes already sitting in your graph - you might need to refresh the node definitions or restart ComfyUI for the new count to fully show up. Which is why, for anything permanent, the env var is the better tool. Set it before startup and every node comes up with your preferred count:

RAT_REMOTE_RUN_NUM_OUTPUTS=20

on the machine running ComfyUI, or add it to your launch script. The node is for when you want to experiment mid-session without a restart; the env var is for when you know you'll always want more sockets. Everything else about this pack's setup is the usual story: install on both machines via Manager or git clone https://github.com/LatentRat/comfy_remote_run into custom_nodes, dependencies are just websockets and requests, no models to download.