Model Passthrough

The Model Passthrough is a clean relay for the denoising model itself - the MODEL object that comes out of a checkpoint or model loader and feeds the sampler. One optional model socket in, the same model out. It doesn't sample, it doesn't merge, it doesn't touch weights. Its entire job is to let the model travel across a big graph without a heavy wire strung over everything else, and to keep a single loader as the one authoritative source.

Where it genuinely helps: subgraphs. When you collapse a generation region into a reusable node, the model is one of the first things that has to cross the boundary - and since everything downstream is shaped by which model you loaded, having it enter on an explicit wire you can see and verify beats having it appear from nowhere. It's also a good citizen in the "one loader, many consumers" pattern from the plumbing playbook: load once, fan out to a sampler, a detailer, and a second pass, all through one passthrough so there's exactly one place to check what model is actually on the wire.

One thing worth knowing that this node does not do: it doesn't keep a LoRA'd model separate from the base. If you apply a LoRA and pass the result through this node, you get the LoRA'd model out - it's a straight relay, which is exactly what you'd expect.

How it works

The simple passthroughs share one generated implementation; this one's run function is return (kwargs.get("model"),). Input in, unchanged out. Same two behaviors as its siblings:

• Unconnected means None. Empty socket, or muted/bypassed upstream, and the output is None.
• Connect-only. No widget - the model has to come from a loader node.

Inputs and outputs

• model (MODEL) in → model (MODEL) out.

One pair. Output feeds any MODEL socket - KSampler, ModelMerge, LoRA apply, anything. It's under Tojioo Passthrough → Simple Passthrough.

Install

Manager → search "Tojioo Passthrough" → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Tojioo/tojioo_passthrough.git

No models, no pip deps. Small single-maintainer pack, GPL-3.0.

Common issues

The None gotcha applies as always: mute the loader upstream and the sampler downstream loses its model with no error from this node. And there's a subtle trap that's easy to miss - because it's a silent relay, if you wire it to the pre-LoRA model by mistake, your whole generation quietly runs on base weights and you'll wonder why your LoRA "isn't working." That's not a bug in the node; it's the cost of having only one wire to check. Verify what's on that wire before you run a long batch.