Resolution Picker

A dropdown that finally knows the right sizes

ComfyUI has a dirty secret: for all its power, picking a resolution is usually a text box you type into from memory. Resolution Picker fixes exactly that one annoyance and nothing else. It's a single node that hands you an organized dropdown of sensible resolutions - grouped by aspect ratio, sorted by size within each group, and pre-loaded with the native sizes of the video models people actually run - and outputs width and height as plain integers for the rest of your workflow.

That sounds trivial until you've tried to remember Wan 2.2's native size at 2am. It's 832x480 or 1024x624, not 960x540. Close but wrong, and wrong matters.

Why this is worth having

The dropdown is curated around a real problem: every model has a native resolution it was trained on, and generating off-native is how you get stretched bodies, duplicated anatomy, and weird artifacts. SDXL was trained on 1024x1024 plus a small set of ratios like 1216x832. SD 1.5 wants 512. Wan 2.2 generates natively at 832x480 or 1024x624. LTX Video is happiest between 512x288 and 1080p. This node bakes those answers into the menu, so you pick "Wan 2.2 Video - 832x480" instead of guessing and crossing your fingers.

How it works

The node is pure plumbing. It reads the selected dropdown string (something like 16:9 Landscape - 1280x720 (HD)), splits off the width and height, and returns them as two INT outputs. Flip the swap_w_h toggle and it swaps them - one click to go landscape-to-portrait instead of hunting for the matching portrait entry.

Two things worth knowing about the mechanism. First, it's built on ComfyUI's newer V3 node interface (comfy_api.latest), which is why the README demands ComfyUI >= 0.18.x and Python >= 3.10. Second, it has zero dependencies and does no math at all - no GPU, no models, no downloads. It literally parses a string and hands back two numbers.

The inputs and outputs that matter

Only two inputs, and you'll set one of them:

• resolution - the dropdown. Defaults to 1:1 Square - 1024x1024. Runs from 512x512 square through 21:9 ultrawide, with dedicated Wan 2.2 and LTX Video sections for the video crowd.
• swap_w_h - a toggle, default off. Swaps width and height.

Outputs:

• width (INT)
• height (INT)

Wire those into anything that accepts a width or height. In modern ComfyUI you can right-click a node like Empty Latent Image and pick "Convert widget to input", then connect these outputs - and your whole latent size now changes by choosing a dropdown entry instead of typing.

Installing it

No dependencies, no model downloads, no checkpoints to fetch. Either search "Resolution Picker" in ComfyUI Manager (pack title comfyui-resolution-picker), or do it by hand:

cd /path/to/ComfyUI/custom_nodes/
git clone https://github.com/cerridan/ComfyUI_ResolutionPicker.git

Then restart ComfyUI. That's the whole install.

Gotchas

• Node doesn't show up? Your ComfyUI is too old. The V3 interface is the one real requirement. On an old install or a stale fork this node silently won't load - update ComfyUI rather than reinstalling the pack.
• The list is closed. It's a COMBO dropdown, so you can't type a custom size. If you need something not listed (say, SDXL's 1216x832), this node won't do it. That said, it's trivially extendable: the resolutions are a plain Python list in src/resolution_picker/nodes.py, so adding your own entry is a two-line edit if you're comfortable poking at source.

If you generate at one resolution forever, you don't need this. If your workflows wander between portrait, landscape, and video models, it's a rare utility node that actually earns its keep.