Common SDXL Resolutions

If you've ever typed 1024x1024 into Empty Latent Image, nodded, then absentmindedly changed it to something like 1920x1080 because that's the aspect ratio you actually needed - and gotten a subject with two heads for your trouble - this node exists so you stop doing that. Common SDXL Resolutions is a dropdown of the nine aspect ratios SDXL was actually trained on. Picking from it instead of typing free-form numbers is one of the easiest ways to stop fighting your own model.

Why it matters. SDXL wasn't trained on arbitrary sizes - Stability bucketed the training set into a fixed set of aspect ratios, all sized to roughly one megapixel: square at 1024x1024, and eight portrait/landscape pairs stepping out as far as 640x1536. Ask for anything outside that set - 1920x1080 straight up, say - and the model does its best on a resolution it's never actually seen, which is exactly where the classic "off-resolution" failures come from: duplicated limbs, a second head growing out of a shoulder, tiled or repeating backgrounds. This node just hardcodes that list of nine as a combo box so you can't typo your way into one of those failures.

Inputs and outputs. One required input, dimensions - a dropdown with entries like "1024 x 1024 (Square)", "832 x 1216 (Portrait)", "1536 x 640 (Landscape)": all nine SDXL-trained ratios, human-labeled so you don't have to remember which is portrait and which is landscape. Pick one and the node splits it into two integer outputs, width and height. Wire those straight into Empty Latent Image, or - if you're also running this pack's Generation Parameters node - convert its image_width/image_height widgets to inputs and feed them from here instead of typing the numbers twice.

Installing it. It ships as part of TW-CUI-Util, a small ComfyUI utility pack whose own README describes it as a grab-bag of ease-of-use nodes for "combining a number of nodes necessary for a workflow into less." Easiest path: open ComfyUI Manager, search "TW-CUI-Util", install, restart. Manual route: cd ComfyUI/custom_nodes && git clone https://github.com/TW-CUI/TW-CUI-Util, then restart ComfyUI. There's nothing to download beyond the node code itself - no models, no extra Python packages called out anywhere in the pack - so this is about as low-risk an install as custom nodes get.

Troubleshooting. Because it's a pure Python dropdown with no external dependencies, there isn't much that can go wrong with the node itself. If it doesn't show up after install, that's a standard "did the pack actually load" problem - check the ComfyUI console on startup for an import error mentioning TW-CUI-Util, and if Manager shows it installed but the node's missing from the search box, a full restart (not just a browser refresh) usually clears it.

Worth knowing if you're newer to this: those nine ratios are strictly an SDXL thing. Flux, Z-Image, and most of the newer flow-matching generation of models take a much wider range of resolutions in the same rough megapixel ballpark without the hard duplicated-anatomy failure - the "generate at these exact buckets or suffer" rule has largely relaxed for them. So don't reach for this node on a Flux workflow expecting the same protection; it's SDXL-specific by design, which is honestly the point. It's not trying to be a general resolution picker, it's trying to stop you from generating SDXL at a size SDXL was never trained on.

One mildly annoying limitation: there's no custom or free-entry option here, so if you actually want an off-ratio image on purpose - cropping to 16:9 after generating at the closest trained ratio, for instance - you'll still want a manual width/height pair for that pass, or just fall back to the stock Empty Latent Image node. This one is for getting your base generation right, not for every latent you'll ever create.