Run Log Pixaroma
Ten run times that tell you if a change actually sped things up
Run Log Pixaroma is the companion to Run Timer with a memory. Every time you press Run it times the whole workflow and drops the finished time onto a list on the node, newest first. The most recent run is highlighted in orange, the quickest of the ten in green. That's the entire pitch, and it's quietly the most useful thing for the "did that change actually help?" question.
Here's the scenario it's built for: you tweak a sampler, swap a LoRA, add an upscale pass, and you think the workflow got faster - but you have no memory of what the last run actually took. Ten runs visible at once lets you watch a workflow get faster over a session, or catch the change that slowed it down by a factor of three. It's the difference between vibes and data.
What it is (and isn't)
It shows only the times, and the list belongs to that one workflow: it saves with the workflow and travels with it, so a different workflow keeps its own separate list. There are no inputs and no outputs - just drop it on the canvas, no wiring needed, and it times every run automatically. Two subtle buttons in the corner export the list as a .txt file or clear it; the same actions plus Copy are on the right-click menu.
Install
It's part of the Pixaroma pack: ComfyUI Manager → search Pixaroma → install → restart, or:
cd ComfyUI/custom_nodes
git clone https://gitlab.com/pixaroma/comfyui-pixaroma.git
No models, no dependencies, no ffmpeg - it's a pure frontend timer with a tiny backend. The pack-wide gotcha applies: hard-refresh the browser tab (Ctrl+Shift+R / Cmd+Shift+R) after an update, since the browser caches old node visuals and that's the usual "it looks broken" cause.
One thing to know before you treat the numbers as gospel: timing a single run on a shared GPU is noisy - other tabs, other queued jobs, and thermal throttling all move the needle. Don't conclude anything from one run; that's exactly why the list of ten exists. If you really want to benchmark, run a fixed-seed workflow a few times and compare the median, not the fluke-fast one.
Inputs (0)
No inputs
Outputs (0)
No outputs