Output Tester - klinter

Every ComfyUI user eventually hits the same wall: a node upstream isn't doing what you think it's doing, and you can't see what's actually coming out of it. Output Tester - klinter is the diagnostic tool for that moment. Plug any connection into it, run the graph, and it tells you - in detail - what type the value really is and what it actually contains. Think of it as a print-debug for your node graph.

The input_value input accepts any type (it's a wildcard), with an optional second optional_input on top. The node then analyzes both and returns three things: type_info (a STRING describing the Python type, whether it's a tensor, its dtype/device, and a guess at the ComfyUI type based on shape - IMAGE vs MASK vs LATENT vs CONDITIONING), value_info (a STRING with the actual contents - tensor shape, min/max/mean, list contents, string value, dict keys), and passthrough, which hands the original input right back out so you can keep the data flowing to its real destination even while you're testing.

That last output is the quietly smart part. You don't have to rewire your workflow to debug: drop Output Tester on a wire, run once, read the console or the outputs, and the data still reaches wherever it was going. The node also prints everything to ComfyUI's console with nice separators, which is the more reliable place to read it when the outputs are enormous.

Where this pays for itself: tracking down silent failures. Is that "IMAGE" actually a 4-channel tensor? Is your conditioning really a list of 2-tuples? Is that API string coming through with escaped newlines? One run answers all of it instead of three speculative fixes. It won't fix anything itself - that's the point - but it turns debugging from guesswork into reading.

Install is the pack standard: search "Klinter_nodes" in ComfyUI Manager, or cd ComfyUI/custom_nodes && git clone https://github.com/klinter007/klinter_nodes then restart. No dependencies beyond what ComfyUI ships. As always with this pack, keep ComfyUI itself updated.

If you only add one debugging habit this year, make it "when in doubt, run it through Output Tester." It's the cheapest way to stop guessing what your graph is doing.