ZeroClip Conditioning Info

Every ZeroCLIP workflow eventually produces a CONDITIONING tensor you can't see - and when the image comes out wrong, you'll wonder whether the conditioning is even what you think it is. ZeroClip Conditioning Info is the debug node that answers that: it prints the shape, dtype, norm, mean, std, and range of any conditioning you feed it, right in the ComfyUI UI.

It's marked as an output node, so unlike most nodes it displays its result on the canvas instead of just passing it along. You don't wire anything downstream of it - you attach it to a conditioning wire purely to look at what's flowing through. If you've ever stared at a KSampler shape error and wished you could see the tensor, this is the fix.

What you'll actually read in its output

• Shape [B, S, D] - this is the one that matters most. B is the batch size, S is the sequence length (should be 77), and D is the embedding dimension. The dimension is where shape mismatches live: if you built a 768-dim ZeroCLIP conditioning but your checkpoint wants 2048, this line tells you instantly.
• Norm - ZeroCLIP vectors live on the unit sphere, so a healthy norm should be close to 1.0. If you're seeing 0.3 or 2.0, something upstream (like a blend that didn't renormalize) has gone sideways.
• Mean, std, range - quick sanity stats. A zero-vector (the Empty Conditioning node) will show mean/std/range all at 0, which is a nice way to confirm your "empty" negative really is empty.
• Pooled line - only present for SDXL-format conditioning; shows the pooled_output shape and norm. If you expected SDXL and don't see this line, the sdxl flag was probably missed upstream.

The info output socket carries the same text as a STRING, so you can pass it to a text-display node if you want it in a different spot - though honestly, the built-in display is all most people need.

When to reach for it

Two moments. First, when a workflow errors on a shape mismatch and you're not sure which node is emitting the wrong size - drop it on the suspect wire and read the dimension. Second, when you're experimenting with the Conditioning Blend node and want to confirm the blended norm stayed at ~1.0. It's the kind of node that feels useless until the first time it saves you twenty minutes.

Install

Part of the ComfyUI-ZeroCLIP-nodes pack. ComfyUI Manager (search "ZeroCLIP"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/ComfyUI-ZeroCLIP-nodes

Restart ComfyUI; it's under ZeroClip/. No pip install, no model files - it's pure inspection.

One honest caveat

It reports the conditioning tensor, not the image. A norm of 1.0 and the right shape doesn't mean the output will be good - ZeroCLIP's whole premise is that some seed vectors resonate with the model and others don't. Use this node to confirm your plumbing is right, then judge the result by eye, not by these stats.