Mpi Int

Mpi Int is the most boring node in the pack, and that's a compliment. It takes an integer in and passes it straight through - an input int, an output int, nothing else, no magic. What makes it worth having is what value nodes do to a workflow: they break a number out of a widget and turn it into a wire that one source can fan out to every consumer.

Here's the pattern. Your KSampler seed, your width, your steps - each lives in its own little box, and each box you have to edit separately. Change the steps in one sampler and forget the other, and you've got a silent mismatch that produces output that's subtly different from what you meant. The fix is the "Convert widget to input" mechanic: right-click the widget on each node, convert it to a socket, then wire all of them to a single Mpi Int. Now there's one authoritative number on the canvas, and you change it in exactly one place. This is the plumbing layer's whole job, and the pack authors know it - they ship typed value nodes (MpiInt, MpiFloat, MpiString, MpiText) precisely because a fixed, explicit type is one less thing to guess about than ComfyUI's old polymorphic Primitive node.

You can also feed Mpi Int from upstream rather than typing: run an MpiMath expression or an MpiListCount into it, and the result becomes a named, wireable integer that anything downstream can consume. In a host-app workflow - and remember this pack is the engine under the Cubric Vision desktop app - that's how a tool passes numbers around without hardcoding.

The one input to care about is int itself, a plain INT field with the usual huge min/max range. There are no other parameters, no outputs to miswire. If the value you feed it is a float, it won't coerce - this is a pass-through for INTs, so convert floats explicitly (MpiConvert handles that) if you're in doubt about types.

Install is trivial: search "ComfyUi-MpiNodes" in ComfyUI Manager and install, or

cd ComfyUI/custom_nodes
git clone https://github.com/MadPonyInteractive/ComfyUi-MpiNodes

then restart ComfyUI. No dependencies, no models. It's one of over a hundred utilities from Mad Pony Interactive.

Honestly, if you're a beginner, the fastest way to feel the point is to wire Mpi Int into a KSampler's seed widget and let the randomize control live on the node instead. When you want to lock a seed you liked, it's now a single field on the canvas instead of a search through the sampler. That's the entire value proposition, and it's enough.