HT Inspector

You've got a workflow that half-works and you genuinely don't know why. Every node says it succeeded, but the image comes out black, or the KSampler is rejecting the conditioning, or some custom node is yelling about a tensor shape you've never seen. HT Inspector is the debug tool you drop into that situation: it takes literally anything - images, masks, conditioning, numbers, a plain string - and tells you exactly what it is, while passing it through untouched.

It comes from HommageTools (ArtHommage), a grab-bag utility pack that's perpetually "alpha" per its own README. Don't let that scare you off; the debug nodes in it are the most battle-tested parts of the pack because everyone who installs it ends up using them.

What it actually does

The node has a single input, data (type *, meaning any type), and two outputs: data (the original value, passed through byte-for-byte) and info (a STRING with everything it learned). Wire the data output to wherever the value was headed next and the info output into a text display node like Show Text, and you get a live readout in the graph instead of guessing.

What's in that readout depends on what you feed it:

• Tensors (images, masks, latents): shape, dtype, device, memory footprint, and a check for BHWC format - batch/height/width/channels. That's the format ComfyUI images live in, and the number one cause of "shape mismatch" errors is accidentally handing a BCHW tensor to something that expects BHWC.
• Lists, dicts, tuples, numpy arrays, strings: type name plus structure details and values where it can print them.
• Anything generic: type name and object size.

One genuinely handy behavior: if you leave the info output unconnected, it prints the full report to the ComfyUI console instead. So you can drop it in, run once, read the terminal, and pull it out - no text node required.

How to use it

The classic debug pattern is to put it on a wire right before whatever's failing. Say your VAE decode output is the wrong shape: insert HT Inspector between the decode and the next node, run, and read the shape. If it says BHWC detected: Batch=1, H=1024, W=1024, C=4, you've got a latent, not an image - mystery solved.

Installing it

You need the whole HommageTools pack - there's no standalone HT Inspector. Easiest path is ComfyUI Manager: search HommageTools and install. Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/ArtHommage/HommageTools.git
cd HommageTools
pip install -r requirements.txt

Then restart ComfyUI. The pack's dependency check wants torch ≥ 2.0, numpy, Pillow, and tifffile, and the requirements file adds psd-tools, dynamicprompts, and google-generativeai (some of which only other nodes in the pack need - you're pulling them anyway).

Gotchas

• It only inspects the first item of a batch tensor in detail, so if you're hunting a per-image problem in a batch of 16, look at the batch size field first.
• The info string is verbose and timestamped. It's built for a quick peek, not for permanent placement in a production workflow - which is fine, because nothing in this pack is meant for production per the author's own disclaimer.

It's not glamorous, but when a workflow is lying to you, HT Inspector is the fastest way to make it tell the truth.