Queue on remote (start of chain)

RemoteQueueSimple-Nux sends whatever's currently loaded in your local graph to a second GPU - simple, but it means your remote job is always tied to your local canvas. RemoteChainStart-Nux is the "advanced" alternative the pack's README describes: instead of the current graph, it takes a full workflow as a JSON input and dispatches that wholesale, independent of whatever you happen to have open locally.

Where it fits

This is built to pair with LoadWorkflowJSON-Nux. Save a workflow once - as an ordinary ComfyUI PNG carrying its embedded graph - load it back with that node, and wire its Workflow JSON output straight into this node's workflow input. From there this node dispatches that fixed, known-good pipeline to a remote instance without you needing to rebuild it in your local canvas first.

The README frames this as the setup for people running more than two GPUs, where you want a farm of remote workers all capable of running the same saved workflow on demand, rather than being limited to the "final_image" convention RemoteQueueSimple-Nux and its FetchRemote-style counterpart depend on.

Inputs and outputs

Required: workflow (a JSON input - feed it from LoadWorkflowJSON-Nux), trigger (on_change re-fires only when inputs change; always forces every run - flip to always if a tweak like a sampler setting doesn't seem to be triggering a re-queue), batch (1–8, default 1), and seed.

Optional: the same ten remoteapply1–remoteapply10 sockets as RemoteQueueSimple-Nux, for wiring in RemoteApplyValues-Nux / RemoteApplyValuesMulti-Nux overrides that patch individual values inside the loaded workflow before it's sent.

Output: remote_chain, a REMCHAIN type - distinct from RemoteQueueSimple-Nux's remote_info (REMINFO). The "start of chain" naming implies this is meant to feed further downstream dispatch logic, though the node(s) that actually consume a REMCHAIN aren't among this pack's more common ones - if you're building around this output, expect to dig a bit further into the pack's advanced queueing nodes than the basic dual-GPU setup requires.

Installing it

Search ComfyUI_NetDist_Plus in ComfyUI Manager. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/nux1111/ComfyUI_NetDist_Plus
pip install requests

Restart ComfyUI. No model downloads for this node.

Where people get burned

Feed workflow a genuinely valid workflow JSON - pair it with LoadWorkflowJSON-Nux rather than trying to hand-type or paste one together.

The classic ComfyUI dependency problem applies here too, just one layer removed: if the saved workflow uses custom nodes the remote instance doesn't have installed, the dispatch will fail on that remote machine, not on yours - so you won't get a clean local error. Check the remote instance's own console log first if a dispatched job seems to go nowhere.

Worth knowing before you invest time here: the pack's own author is upfront that editing a loaded workflow isn't built yet - no search-and-replace on a seed or swapping a checkpoint name inside the JSON itself. That's exactly the gap RemoteApplyValues-Nux and its Multi sibling are meant to paper over, one field at a time, via the remoteapply slots. If you need to change more than a couple of values, expect to be stacking several of those override nodes rather than editing the workflow directly.