LatentPass

A node that passes its input straight through to its output sounds useless. In a small graph, it is. In the kind of giant modular workflow ComfyUI Flux Continuum is built around, a deliberate wire-breaker like this is quietly load-bearing. LatentPass exists so your latent can keep flowing through a section of the graph without being consumed, renamed, or lost in the spaghetti.

What it actually is

One input, one output, both LATENT. Feed it the latent from a KSampler, a VAE encode, or a latent-to-latent upscaler and you get the exact same object back. The Python is three lines - take the latent, return the latent - so there's no transformation, no side effects, nothing to tune.

That "same object" bit matters more than it sounds. ComfyUI passes tensors around by reference, not by copy. So LatentPass isn't a fork or a snapshot: if the upstream node that produced the latent reruns with different settings, whatever you pulled out of LatentPass changes too. It's a wire, not a bucket.

Why you'd bother

Flux Continuum is built on a get/set routing system - values get "set" into named slots and pulled out elsewhere, which is how its one-interface-controls-everything trick works. Pass nodes are the other half of that design: they keep a data type flowing through a branch of the graph while the set/get machinery does the rerouting around it. The pack ships a whole family - Latent, Pipe, SEGS, and Int passes - "to maintain data flow integrity," as the README puts it.

Outside the Continuum workflow you'd reach for it for the same reasons people use Reroute nodes: to break up an unreadable wire, to keep a clean copy of the pre-sampler latent dangling out of a branch you're testing, or to have a stable place to grab the latent when half the graph around it gets bypassed (Ctrl+B). It's the node you add when the alternative is dragging a connection across the whole canvas.

Install

It ships with the whole pack, so install once:

cd ComfyUI/custom_nodes
git clone https://github.com/robertvoy/ComfyUI-Flux-Continuum

Then restart ComfyUI. Easier: open ComfyUI Manager, search "Flux Continuum", install, restart. The pack has no Python dependencies of its own (pyproject.toml declares none), though the full Continuum workflow leans on Impact Pack, rgthree, Essentials, and KJNodes - Manager will offer those when you import the workflow.

Troubleshooting

Honestly, the failure modes here are boring. If the latent never reaches your KSampler or VAE Decode, check the wire is actually connected to the output socket - a passthrough node can't corrupt anything. The one thing people get wrong is expecting it to freeze a value: bypass a node upstream and LatentPass hands through whatever comes next. If you want a true copy, use the node that actually re-encodes or clones the latent, not this one.

If ComfyUI reports the node as missing on workflow import, the pack didn't install cleanly - reinstall via Manager, and don't skip the dependency packs it lists. That's the whole job. Sometimes the most useful node in a workflow is the one that never does anything.