Run Timer 🆎

You changed one LoRA strength and the run felt slower. Was it? Run Timer 🆎 answers that with a black readout you drop anywhere on the canvas and then ignore until you care.

It has no inputs and no outputs. That's not a stripped-down version of the node - that's the whole node. It doesn't sit on a wire, so it can't break a graph, can't be bypassed by mistake, and can't affect execution order. It listens to the queue instead.

How it works

The clock starts when a queued prompt begins executing and ticks while it runs; when the run finishes the total holds. The state is a dot colour: amber while running, teal when held, red for a failed or interrupted run - and a failed time is deliberately not kept as a reference, which is the one opinionated choice here and the right one.

Two boundaries are worth knowing precisely, because they're the difference between a stopwatch and a benchmarking tool:

• It measures from execution_start to the end of the run, on the browser's clock. Model loading counts. Time spent sitting in the queue behind another prompt does not.
• It is frontend-only. Nothing about it reaches the API prompt, so a headless run won't report a time.

The last completed run is saved with the workflow. That's the quietly clever part for anyone who downloads and shares graphs: a shared workflow reopens showing its author's run time, which turns "should this take 40 seconds or four minutes" into a question with an answer. Hover the box for the previous few, or right-click for the list. Right-click → Reset Run Timer history clears it.

Because the node is literally just the box - painted by the node itself on the classic canvas, no title bar, no pack badge - you drag it from anywhere and resize it by the corner, and the digits scale with it. Put it in the corner of a heavy video workflow and leave it there.

What it doesn't tell you

If you want to know which node is the slow one, this isn't that tool. That's a separate feature in the same pack: Settings → 🆎 AusBoss → Chrome → Node runtime badges stamps each node with its own seconds. Between the two you get "the run took 92 seconds" and "71 of those were the sampler", which is the pair that actually lets you optimise something.

A couple of small things that trip people:

• Several Run Timer nodes on one canvas all show the same run. It's measuring the queue, not itself.
• The elapsed total is per-run, not per-session. It resets on the next queue rather than accumulating.
• If you want an audible cue when the queue empties, that's Settings → 🆎 AusBoss → Notifications → Completion sound - not this node.

Installing it

ComfyUI Manager → search AusBoss, or:

cd ComfyUI/custom_nodes
git clone https://github.com/ausboss/ComfyUI-AusBoss.git

Restart ComfyUI and hard-refresh with Ctrl+Shift+R. Since the entire node is a frontend drawing of queue events, a stale cached frontend is how you'd get a blank square where the timer should be - the refresh is the fix, not a reinstall. No models, no pip extras, and the pack needs ComfyUI 0.27.1 or newer.

One caveat on comparing numbers you see on the internet: these timings are wall-clock on someone else's machine. Same graph, same settings, different GPU - the graph is reproducible, the timing isn't. Your own previous run is the only benchmark worth reading.