输出目录传递

OutputDirPassthrough takes a string in and gives you the exact same string back. That's it. It's the pack's version of a pass-through wire, and it exists for a boring but real ComfyUI reason: sometimes a node produces a value that you can't cleanly re-feed into another node's input, and you need a typed adapter in between.

Here the value is output_dir. GeneralConfig computes where your training run's outputs will live (inside ComfyUI's output/ folder) and hands it out as one of its three outputs. TensorBoardMonitor needs that same path as a plain string input so it can point tensorboard --logdir at it. The passthrough node sits in between, converting the output slot into an input slot cleanly.

How it works

One input, output_dir (a STRING marked forceInput - meaning ComfyUI always shows it as a wire input rather than a text box), one output output_dir (STRING). The passthrough function literally returns the value unchanged and prints a little log line. It doesn't check that the path exists, doesn't touch WSL path normalization, doesn't do anything clever.

In the example workflow it's wired like this:

GeneralConfig (output_dir) → OutputDirPassthrough → TensorBoardMonitor (output_dir)

You can think of it as documentation as much as plumbing: it names the data flowing from the training config into the monitor, which keeps the graph readable.

The input that matters

• output_dir - the training output directory string, from GeneralConfig's output_dir output. The output of this node carries the same string onward.

That's the whole interface. There is genuinely nothing else to configure.

Install

It ships with the pack, so install the pack:

cd ComfyUI/custom_nodes
git clone --recurse-submodules https://github.com/TianDongL/Diffusion_pipe_in_ComfyUI.git
cd Diffusion_pipe_in_ComfyUI
git submodule init && git submodule update
pip install -r requirements.txt

Or use ComfyUI Manager and search "Diffusion_pipe_in_ComfyUI". Linux/WSL2 only, as with everything in this pack.

Where people get burned

Honestly, the only way to get burned here is to skip it and wire GeneralConfig's output_dir directly into TensorBoardMonitor, then hit a type mismatch or have the field render as a text box instead of a wire. If that happens, just drop this node back in. It's also worth knowing it does zero path validation, so if your training logs end up nowhere sensible, the problem is upstream in GeneralConfig, not here. This is the rare ComfyUI node where "it does nothing" is the correct behavior.