Sampler Pipe In

Sampler Pipe In exists to make wide graphs smaller. You've got a model, positive and negative conditioning, a VAE, maybe an image, a clip, a latent - seven wires snaking across the canvas into your sampler. This node bundles as many of them as you care to feed into a single pipe output, so one wire does the work of seven.

The idea is familiar if you've used rgthree's Context nodes or Impact Pack's pipe machinery - the community has converged on "pack related stuff into a bundle, pass one wire." This is the sampler-flavored version: every input is optional, and anything you leave disconnected is simply left out of the bundle.

How it works

The mechanism is boring in the best way. Each input that's connected gets dropped into a plain dictionary under its name - model, positive, negative, vae, image, clip, latent - and that dict is what comes out of the pipe output. Disconnected inputs don't appear in the dict at all; there are no phantom empty values wandering around.

Inputs and outputs

All inputs are optional: model (MODEL), positive (CONDITIONING), negative (CONDITIONING), vae (VAE), image (IMAGE), clip (CLIP), latent (LATENT). Output is a single pipe (*).

The natural use is this pack's Sampler Pipe Out node on the other end, which unpacks the same keys back into individual outputs. Pack once, split wherever you need it - including splitting into two places at once, which is a genuinely nice trick for branching a pipeline without re-creating every connection.

The catch

The pipe is an untyped dictionary. That's what makes it flexible, but it also means it only reliably talks to this pack's own pipe nodes. Don't expect to plug it into Impact Pack's pipe, rgthree's Context, or anyone else's - those are different bundles with different schemas, and ComfyUI won't stop you from trying because the type is *. It'll only fail at runtime, which is exactly the kind of failure that's hard to spot in a big graph.

One more thing to like about the bundle approach: because missing inputs are just absent, you can reuse one pipe for two different sampler configs without building weird conditional wiring. If a key isn't there, the unpack side gets nothing for it.

Installation

Part of the ComfyUI-Jax-Nodes pack. ComfyUI Manager: search "ComfyUI-Jax-Nodes", install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Kebolder/ComfyUI-Jax-Nodes

No pip dependencies, no models to fetch. The pack targets ComfyUI's newer extension API, so keep ComfyUI on a current build (late 2025 or later). Lightly maintained, small pack - but a bundler node is about the most stable thing you can ship.