Aspect Ratio JK🐉

Every diffusion model has resolutions it was trained at, and feeding it something way off that range is how you get doubled heads and cropped compositions. Aspect Ratio JK🐉 is a 31-entry preset menu that hands you a width and height in the native shape of the model you're running - no mental math about what SDXL's base resolution should be.

It's a small utility, but it's the kind of thing that quietly unblocks beginners. Instead of typing 1152×896 from memory every time you build an SDXL workflow, you pick "SDXL 1152x896" from the list and wire the two INT outputs wherever a resolution is expected: Empty Latent Image, Empty SD3 Latent, or the width/height inputs on a KSampler-parameters node.

How it works

The mechanism is boring in the best way. A dropdown maps a friendly label to a hardcoded resolution, the node parses "SDXL 1152x896" into two ints, and returns them. The presets are grouped by the model family they belong to - SD15 has a whole band of 512-tall landscape shapes (680, 768, 912, 952, 1024, 1224 wide), SDXL gets its 1024-class portrait/landscape set, SD3 adds a few of its own, and Qwen-image has 1328×1328 and friends. The custom path is where it gets more useful than most aspect-ratio nodes: set resolution to "Custom" and the custom_width / custom_height sliders (64–16384, stepping by 8) take over.

swap_dimensions is the one toggle that will save you an afternoon. Set it true and the node returns height, width - you can keep your landscape preset while composing vertically, or rotate a portrait shot for a social-media crop without changing the preset.

The inputs that actually matter

• resolution - the preset list. Start here.
• custom_width / custom_height - used only when resolution is "Custom".
• swap_dimensions - flips the output pair.

Outputs: width and height, both INT. That's it. There's a small guard in the source - if the resolution string somehow fails to parse, it falls back to 512×512 rather than crashing the graph, which is a nicer failure mode than a red node.

Install & gotchas

Part of the JakeUpgrade pack, so: ComfyUI Manager → search ComfyUI-JakeUpgrade, or

cd ComfyUI/custom_nodes
git clone https://github.com/jakechai/ComfyUI-JakeUpgrade
cd ComfyUI-JakeUpgrade && pip install -r requirements.txt

No models, no heavy deps beyond the pack's standard list (opencv-python, pillow, torch). One honest note: presets are a convenience, not a rule. SDXL happily generates at non-native sizes when you denoise properly or upscale after; the preset just removes the "which number do I type" step for the common cases. And if the node doesn't show up in your library, check the pack's config.ini - deprecated nodes are hidden by default, though this one isn't among them, so it should just be there.