HT Tensor Info

"Shape mismatch" is the single most common error in ComfyUI, and the error message rarely tells you the actual shape of the tensor that offended it. HT Tensor Info (from the HommageTools pack) is the debug node that fills that gap: feed it any image tensor and it reports the shape, the layout (BHWC vs HWC), dimensions, batch size, and channel count - and, crucially, it passes your image through untouched, so you can drop it into an existing wire and inspect without rebuilding anything.

How it works

One input, two outputs:

• image - the tensor to inspect.
• image (output) - the exact same tensor, passed through. Nothing is modified, no copies that change your data.
• shape_info - a string describing what was found.

Internally it reads tensor.shape and distinguishes the two layouts that matter in ComfyUI: a 4-dimensional tensor is BHWC (batch, height, width, channels) - the standard IMAGE format - while a 3-dimensional one is HWC (a single image without the batch dimension). It reports the resolution, batch count, and channel count in a short summary. The summary also gets pushed to the UI on the node itself via ComfyUI's preview-update mechanism, so you see the info without hunting for a text output - though shape_info is there to wire into a display node if you want it in the graph.

It re-evaluates every execution (IS_CHANGED returns nan), so the shape it shows is the shape from this run, not a stale cache. That matters more than you'd think when you're debugging a workflow where shapes change between runs.

When you'd reach for it

Any time a node rejects your image and you don't know why. Was it the batch size (your upscaler output a batch of 4 when the next node wanted 1)? The channel count (an RGBA image where a node wanted RGB)? The missing batch dimension (an HWC tensor where BHWC was required)? One read of the shape_info and the answer is usually obvious. It's also useful mid-workflow: insert it after any node that transforms dimensions - resize, crop, upscale - and you can verify the transformation actually did what you think it did.

Installing it

HommageTools install, once:

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

Restart ComfyUI, or ComfyUI Manager → "HommageTools". No models, no heavy deps.

Where it falls short

It inspects image tensors only - the input socket is typed IMAGE. If your problem is a latent, a mask, or some other tensor type, this node can't look at it (the pack's HT Inspector is the any-type option for those). And "info" here is shape-focused: it won't tell you about value ranges, device placement, or memory. For the classic "why won't this connect" debugging job, though, it's the first node to drop in - it answers the most common question with the least ceremony.