Comfy Kitchen Dispatch Report (Legacy Video)

Video runs are where telemetry stops being a curiosity. A still takes eight seconds, so guessing is free. A twenty-second clip with a sampler pass, an upscale and an interpolation stage takes real minutes, and "it felt slow" is not a diagnosis. This node is the pack's telemetry report, wired into the video path: display name Comfy Kitchen Dispatch Report (Legacy Video), a required VIDEO input, a VIDEO back out, and the report alongside it.

What it does and how

Mechanically it's the same engine as the rest of ComfyUI-CK-Telemetry: it hooks ComfyUI's comfy_kitchen dispatch registry to count backend selections per operation for the prompt, samples CPU, RAM, pagefile, GPU utilization and VRAM on a 500 ms tick, and formats the result as text. It never touches your frames - the video output is the same object you handed it, passed through unchanged, which is the point of a diagnostic node. Nothing gets re-encoded, no frames get copied, and the graph's output is bit-identical to what it would have been without the node.

The report output is a STRING, so you can send it to Preview as Text if you want it off the node. With embed_metadata on, the structured report and its text form go into EXTRA_PNGINFO for the next saver, and the pack's frontend extension refills the node's read-only report field when a generated file is dropped back onto the canvas. The rest of the controls are the pack's shared set: show_terminal_report and terminal_report_level for the terminal dump, node_report_level (summary or detailed) for the field/output/metadata, and show_node_report to collapse the report field when your graph is already a wall of nodes.

The elapsed figure is wall-clock time measured up to this node. On a video workflow that's the number you want, as long as you place the node after the stages you're judging.

The catch: it needs a real VIDEO type

VIDEO is ComfyUI's native video datatype, and it only exists on builds that have the video nodes. If your setup still represents clips as an IMAGE batch - the older pattern - this socket simply has nothing to offer you: the node's required video input can't be satisfied, and on a build old enough to lack the type in its node API, registration is where things fall over. Update ComfyUI first. If you're deliberately staying on an older build, use CK-Telemetry instead: its wildcard input/output will carry whatever your pipeline is actually passing around.

Also note this node is not an output node, and both video and report are outputs. It runs because a downstream node needs the video - so keep it in the live branch that reaches your save or combine node. A disconnected output means no execution and no report.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Elsewhere101/ComfyUI-CK-Telemetry

Restart, then look for ComfyKitchenTelemetryVideo in the diagnostics/comfy kitchen category. The pack is installable from ComfyUI Manager too - search CK-Telemetry - and its only real dependency is psutil>=5.9.0.

Troubleshooting

• Fields are blank after a render. The node didn't execute. Check nothing is bypassing it and that its video output actually reaches a downstream node.
• "No Comfy Kitchen backend selections observed." Nothing in that prompt went through the dispatch registry - most often an older ComfyUI. Check inside ComfyUI's own environment with python -c "import comfy_kitchen as ck; print(ck.registry.list_backends())".
• Metadata missing from the saved file. embed_metadata needs a native saver downstream, and --disable-metadata on the launch command undoes it regardless of the widget.
• Time looks short. You placed the node early. The measurement stops where the node runs.

It's marked deprecated, and the author's own advice is implicit in that: new graphs should use the wildcard CK-Telemetry node. But if a downloaded video workflow already contains this one, leave it - it works, and it's doing exactly what it says.