Resolution Selector (Advanced)

The "(Advanced)" in this node's name is doing real work: it lets you size by edge or by megapixels rather than by brute-force width/height, and it rounds every result to a clean step so the pair is one a diffusion model will actually accept. If you've ever typed 1344×768 into a loader and wondered why it underperforms, this is the node that stops you doing that by hand.

Resolution Selector answers one question - "what width and height match this shape and budget?" - and answers it in the units you actually think in. Pick an aspect ratio, say how you want to size it, and it returns width, height, the ratio it actually achieved, the megapixel area, and a ready-made 1024 x 576, 16:9, 0.59 MP label for your filename.

How to drive it

• aspect_ratio - the shape, widest side first. 16:9 widescreen, 3:2 stills, 1:1 square. orientation flips it to portrait; square ignores the ratio entirely.
• size_by - which measurement you're giving: long edge, short edge, width, height, or megapixels. 1024 on the long edge of 16:9 gives 1024×576; 1.0 megapixels at 16:9 gives roughly 1344×768.
• size - the pixel measurement, read by every option but megapixels (which reads its own megapixels field instead).
• multiple_of - the step both sides land on. 8 is the least a latent will take; 64 is what most model families were trained on. Set it to match your checkpoint's native multiple and the pair stays "model-safe."

Why the outputs matter

width and height wire into any loader or latent node that takes dimensions. The quieter outputs are the useful ones for automation: ratio tells you the actual width÷height after rounding (compare it to what you asked for and you'll see when the step forced a slightly different shape), megapixels reports the real area so you can keep two different shapes to the same render budget, and label drops a description into a filename prefix or caption so a render remembers its own settings.

The honest caveats

The tooltip flags the real gotcha: 1024 on the long edge of a 16:9 image is 1024×576, which is only 0.59 MP - noticeably softer than a square 1024×1024. Long, thin images look high-res and aren't. If you're comparing renders across aspect ratios, size by megapixels instead, and keep the megapixels output around so the comparison is apples to apples.

Also remember this gives you a pair, not a loaded model - wire width and height into the empty image or latent you're about to sample, don't expect the resolution to be enforced anywhere else.

Installing

Resolution Selector lives in WAS Node Suite v3. Through ComfyUI Manager, search WAS Node Suite v3 and install; manually:

cd ComfyUI/custom_nodes
git clone https://github.com/WASasquatch/was-node-suite-comfyui

Restart ComfyUI. Requirements: ComfyUI 0.14.0+ and Python 3.10+. This node does pure arithmetic, so it's a good introduction to the pack's v3 philosophy: no dependencies, nothing downloaded, no weights. The old v2 WAS suite pulled in a pile of packages and broke on ComfyUI updates; v3 computes on the torch device you already have.