EmAySee Aspect Ratio Calculator

ComfyUI will happily upscale an image into the wrong shape if you hand it bad numbers, and "bad numbers" usually means a width that doesn't match the image's actual aspect ratio. EmAySee Calculate Aspect Ratio is a three-input sanity node: feed it an image and a target height, and it returns the width that keeps the original proportions.

That's the whole job - and it's genuinely the right job for the upscaling half of a workflow. Say you have a 1536×1024 image and you want to feed it to an upscaler at a specific height. You type 1024 as the output height, and the node computes width = height * (original_width / original_height) - so you get back a width that preserves the frame instead of stretching it. It keeps image or latent pipelines honest when the downstream node needs explicit dimensions.

How it works

The mechanism is a single division, visible in py/EmAySee_CalculateAspectRatio.py. It reads the image's shape, computes the aspect ratio, and multiplies by your target height:

ratio = original_width / original_height
output_width = int(output_height * ratio)

The height you provide is returned verbatim; only the width is computed. It's a proportional scaler, not a resizer - it produces dimensions for you to feed elsewhere, and produces no image itself.

The inputs that matter

• image - the reference image whose aspect ratio you want to preserve.
• output_height - the target height, default 512, minimum 1.

Outputs are output_width and output_height, both INT.

Install

Ships in the EmAySee pack:

cd ComfyUI/custom_nodes
git clone https://github.com/EmAySee/ComfyUI_EmAySee_CustomNodes

Restart ComfyUI (or Manager → "ComfyUI_EmAySee_CustomNodes"). Category: EmAySee_Utils.

Gotchas

The trap worth knowing: the result is not rounded to a model-friendly multiple of 8 (or 16, or 64 depending on your model). Your 1536×1024 image scaled to a target height of 1024 gives you width 1536 - clean - but many arbitrary inputs won't come out so tidy, and feeding a non-multiple-of-8 dimension into a latent or a sampler can cause errors or subtle artifacts. Round the output yourself with a math node (int(w / 8) * 8) or use a resolution-aligning node afterward. And the standing pack caveat applies: unsupported hobbyware, no guarantees - trivial math, but verify the output before you pipe it into an expensive generation.