WH Pixaroma

Sometimes a problem is just two numbers. You want to generate at a specific resolution, you want to feed a width and height into an empty latent, and you don't want to drag an aspect-ratio picker or a resizer into the graph just to do it. WH Pixaroma is exactly that: two number fields, two number outputs, nothing else. Type 704 and 1280, and out come width and height ready to wire wherever a size is needed. It's the smallest node in the pack and it does one thing, which is why it never breaks.

Why you'd reach for it

The classic setup is a latent. ComfyUI's Empty Latent Image has its own width/height boxes, sure - but if the size has to come from somewhere else in the graph (a value you computed, a node you toggled, a resolution you want to share between a latent and a resize), you need a numeric source. WH gives you that source, and because it's a pure value node it slots in anywhere numbers go: an empty latent, an upscale, a video frame size, a crop.

Two details make it nicer than a bare pair of ints:

• Math in the fields. 1024+64, 512*2 - type an expression and it evaluates. Handy when you're targeting a model-friendly number off an existing base.
• Pairs with Switch WH Pixaroma. Feed WH's output into one side of a Switch WH and a Load Image's WIDTH/HEIGHT into the other, and you can flip between "typed size" and "whatever the loaded image is" with one click. That's the combo the README explicitly pitches, and it's genuinely useful for switching a workflow between manual and image-driven resolutions.

What's on the node

• width / height inputs - INT fields, both defaulting to 1024, ranging 64 to 16384 in steps of 8. The step size is deliberate: diffusion models generally behave on multiples of 8, and many are fussier still (that's why the pack's other size nodes offer snap-to-8/16/32/64). Type any value you like; the arrows just nudge by 8.
• width / height outputs - INT, exactly what you typed. Wire them into an Empty Latent Image, a Resize Crop, a video latent, wherever.

No inputs to wire, no models, no dependencies. It's the sort of node you'd normally just rebuild with two Number nodes, and WH does it with a third fewer nodes and a size-specific UI.

Install and footnotes

It's part of the Pixaroma pack:

cd ComfyUI/custom_nodes
git clone https://github.com/pixaroma/ComfyUI-Pixaroma

or install via ComfyUI Manager (search Pixaroma) and restart. The pack has no required Python dependencies beyond ComfyUI's own PIL/torch. If the fields ever look stale after a pack update, hard-refresh the browser (Ctrl+Shift+R) - the cache, not the node, is the usual culprit. Also, if you want a one-click orientation flip instead of typing a second pair of numbers, its sibling Portrait Landscape Pixaroma does exactly that; WH is the one to grab when you want full manual control.