Deprecated · Sampling Trace Image

Sampling Trace Image is a probe: you drop it on an IMAGE wire, it records a numeric summary of whatever flows through - shape, dtype, value range, mean, standard deviation - and then hands the image back out completely untouched. It doesn't resize, recolor, or filter anything. If you've ever wired a node in just to eyeball "is this image normalized or is it in the 0–255 range?" this is that question, answered mechanically, mid-graph.

It's part of the pack's hidden "probes" family (with Latent, Mask, Conditioning, and Model Snapshot). The current 0.4.0 release marks all of them deprecated and hides them from normal node search - they exist for saved-workflow compatibility and the author's internal diagnostics. Probe nodes don't start sampling trace on their own; they're observation posts, not the trace itself.

Inputs and outputs

• image (required, IMAGE) - whatever branch you want to inspect.
• label (optional, STRING) - a name so you can tell this probe apart from the other five in the panel. Cheap and worth it.
• trace_session (optional, TRACE_SESSION) - if you wire one in (from the deprecated Sampling Trace Model or the one-node setup's TRACE_SESSION), the summary is stored into that run's evidence. Without it, the probe just pushes a live event the panel can show.
• Output: image, unchanged.

Install

Same pack, same command:

git clone https://github.com/ILcrowe/ComfyUI-SamplingTrace-Inspector.git ComfyUI/custom_nodes/ComfyUI-SamplingTrace-Inspector

or ComfyUI Manager → "Sampling Trace". Restart. No model files, no extra pip packages.

Where people get burned

• "I can't find this node anymore." Correct - probes are hidden from normal search in 0.4.0. They load from saved workflows; new graphs are steered to the one-node setup.
• Expecting it to start tracing. It doesn't. The description says it plainly: probes record and pass through; they don't begin sampling trace capture.
• Not wiring trace_session and wondering where the data went. Without a session, the summary is a live event, not persisted history. If you want it saved with a run, connect the session.

Honest verdict: for most people this node is a museum piece that proves the pack's architecture. If you're debugging a weird image branch in an old workflow, it's genuinely handy - but you'll get further with the one-node setup and the panel's step data than with probes in the wire.