Deprecated · Sampling Trace Latent

Sampling Trace Latent is the probe node for the compressed representation. You drop it on a LATENT wire and it reports what the tensor actually looks like - shape, keys, value range, statistics - then passes the latent through completely untouched. It's the "is my latent blown out or sane?" question, answered numerically in the middle of your graph, with zero risk to the data.

The useful thing to know about LATENT objects: they're not a single tensor, they're a dict with a samples key plus whatever metadata a node stuffed in (noise masks, batch indexes, sampler-specific fields). This probe summarizes the samples tensor and also lists the latent's keys, so you can see at a glance whether a custom node added something unexpected to the dict. That's the kind of visibility that saves you an hour of "why is my img2img acting weird" staring.

It's one of the pack's five hidden probe nodes - deprecated and hidden from normal search in 0.4.0, kept for saved-workflow compatibility and internal diagnostics. It does not start sampling trace on its own.

Inputs and outputs

• latent (required, LATENT) - the branch to inspect.
• label (optional, STRING) - a name for the probe in the panel; helpful when you've got a few of these in one graph.
• trace_session (optional, TRACE_SESSION) - wire one in to store the summary into that run's evidence instead of just emitting a live event.
• Output: latent, unchanged.

Install

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

or ComfyUI Manager → "Sampling Trace", restart. No dependencies beyond ComfyUI's own environment, no model downloads.

Where people get burned

• Hidden from search. Deprecated by design - you'll only meet this node in an old saved workflow. New graphs use the one-node setup.
• Expecting it to capture sampling. It doesn't; probes don't start trace capture. It records and passes through.
• Forgetting trace_session. Without it, the summary is a live event and won't be in your run's stored files. Connect the session if you want history.

Realistic use: most people never need it. But when you're debugging why a latent branch misbehaves in an old workflow, a numeric summary beats squinting at a preview.