Monitor Pixaroma

Monitor Pixaroma is a small dashboard that sits on your canvas and shows what your computer is actually doing while it generates: video memory, system RAM, GPU load, CPU load, GPU temperature, and power draw, all updating live. There's nothing to wire. You drop it on the canvas and it just works. No inputs, no outputs, nothing to configure beyond cosmetics.

The killer feature is the peak mark. A pale line on the video memory bar shows the highest point your VRAM reached during the last run. That one number answers the question everyone asks after a workflow that nearly died: how close to the top did I get, and would a bigger model still fit? Leave one on while you tune a workflow and you'll quickly learn which stage is the hog.

Worth knowing how it works, because it explains the limits. The node never runs in Python - it's a frontend-only control, skipped on every Run, all the work happening in JavaScript that polls a stats endpoint on ComfyUI's own server. The backend reads VRAM straight from torch via ComfyUI's memory management, the same numbers its system stats panel reports, and shells out to nvidia-smi for the extras - utilization, temperature, power - in a background thread with cached results, so the polling never stalls your generation. System RAM and CPU come from psutil, which ships with ComfyUI anyway. No new dependencies, no model downloads.

There's also a Free VRAM button right on the node. It unloads the models ComfyUI is holding - the same thing as the "Free model and node cache" item in ComfyUI's own menu. Handy right before you load a big model, or when another program wants the card. It's the manual version of what the sibling Free VRAM node does automatically at a chosen spot in the graph.

Right-click, or the gear on the toolbar, to pick which readouts to show, switch between the full bars and a one-line strip, and set how often it updates. Drag the corner to make it bigger. The live numbers deliberately aren't saved into your workflow - a node writing a value every second would mark the file modified constantly and fill the undo history with junk - but your layout choices are.

Limits, honestly stated. GPU utilization, temperature and power come from nvidia-smi, so on AMD cards or machines without it those rows quietly drop out; you still get VRAM from torch and CPU/RAM from psutil. And some laptop GPUs don't report power or temperature at all - the node shows N/A rather than a fake zero, which is the right call. If nvidia-smi keeps failing it stops trying for the session, so a restart is the retry.

It's part of the Pixaroma pack - the same suite that brought you Image Composer, Paint and the 3D Builder, from the designer-turned-YouTuber who runs the beginner-friendly Pixaroma tutorial series. Install via ComfyUI Manager (search Pixaroma, install, restart) or clone https://gitlab.com/pixaroma/comfyui-pixaroma.git into custom_nodes/. It landed in the same August 2026 update as Free VRAM, so if it renders oddly after updating, Ctrl+Shift+R hard-refresh the tab before blaming the node.