Tensor Show

Debugging a ComfyUI workflow is miserable in the default setup. Your image comes out wrong and you've got nothing between "click queue" and "look at the PNG" except your own guesswork. TensorShow from the MoneyMaker pack (display name ComfyUI-Yuan) is the pack's answer: a node you drop anywhere in the graph to dump what's actually flowing through that wire - shape and values - straight into the terminal and onto the node itself.

How it works

It's a wildcard input (type *), which means it accepts absolutely anything: an image tensor, a latent, a mask, a conditioning dict, a list, whatever. When the graph runs, it walks whatever you fed it - recursing through dicts and lists, and pulling the .shape off any tensor it finds - then prints two things to ComfyUI's console: the full Tensor values: dump and a Tensor shape: summary. A small frontend extension the pack ships (moneymakerweb/moneymaker.js) also pushes that text into a read-only, monospace, multiline widget on the node, so you can read it without digging through logs. The node is flagged as an output node, and it sits in the pack's MoneyMaker😺/show submenu.

The sibling "Something Show" node does the same trick for plain values - strings, scalars, JSON - so between the two you can inspect just about any intermediate in a workflow.

The one thing you should know

The schema declares a tensor output, which sounds like a passthrough you could wire onward. Don't build on that: the function only ever returns the UI text message and never actually returns the tensor. In practice the output pin hands downstream nodes None. Treat TensorShow as a terminal debug node - put it at the end of a probe branch, read the shapes, delete it, and carry on. If you genuinely need a pass-through debugger, you're looking at WAS or rgthree territory instead.

Also be careful what you feed it. It prints the entire tensor to the console - values included, not just the shape. A 1024×1024 latent is a wall of numbers that'll swamp your terminal and slow the run. Feed it a single small tensor or a shape probe, not a big batch.

When to reach for it

When the error message points at "a tensor of wrong shape" and you can't tell which one. When a node silently eats your conditioning and you want proof. When you're porting someone's workflow and need to confirm what a custom node actually outputs before you trust it. It's a print statement with a face - and for a beginner, seeing shapes laid out on-screen is genuinely the fastest way to build intuition for what these tensors are.

Install

It's part of the MoneyMaker pack - search ComfyUI-Yuan in ComfyUI Manager (that's the pack title), install, restart:

cd ComfyUI/custom_nodes
git clone https://github.com/Cyber-BlackCat/ComfyUI-MoneyMaker

Then restart. No model downloads; the only real additions are opencv-python and scikit-image, which install automatically. Fair warning that this is a small hobbyist pack (category MoneyMaker😺), so keep expectations proportionate - but for "show me what's on this wire," it does exactly the job.