Image Size Selector

Instead of typing two numbers into an empty-latent node and trying to remember which SDXL resolutions are "blessed," you pick from a dropdown of labeled presets and this node hands you the width and height as integers. Small convenience, real time-saver - especially when you're switching between aspect ratios a lot or you can never remember whether it's 832×1216 or 896×1152 for portrait SDXL.

It's a selector node from Jared Therriault's JNodes suite, and it comes with a genuinely long menu: 52 presets ranging from tiny 128×128 up through the common 16:9, 4:3, and 1:1 sizes, each tagged with its aspect ratio right in the label ("720x480 (16:9)", "640x480 (4:3)", and so on). If none of them fit, there's a custom-size escape hatch.

How it works

You choose a preset from the image_size list and the node outputs the corresponding width and height as two INT pins. A few modifiers sit on top: you can flip the dimensions for portrait/landscape, scale everything by a multiplier, or bypass the presets entirely with a custom width and height. The outputs are plain integers, so they plug straight into an EmptyLatentImage (or anything else that wants a width and a height).

The inputs and outputs

• image_size (enum, 52 presets) - the labeled resolution you want.
• flip_width_and_height (default false) - swaps them, so one preset serves both orientations.
• multiplier (default 1.0) - scales both dimensions; handy for a quick hi-res target off a base size.
• use_custom_size (default false) with custom_size_x / custom_size_y (default 512) - ignore the presets and specify your own.
• width and height outputs (both INT) - wire these into your latent.

How to install it

Via ComfyUI Manager: Install Custom Nodes, search "JNodes", install, restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/JaredTherriault/ComfyUI-JNodes
pip install -r ComfyUI-JNodes/requirements.txt

then restart ComfyUI. No model downloads.

Common issues & troubleshooting

Watch the multiplier - it can produce ugly dimensions. Stable Diffusion wants dimensions divisible by 8 (and behaves best on multiples of 64). A multiplier like 1.5 on an odd base can hand you a width that isn't a clean multiple, which some samplers tolerate and some don't. After applying a multiplier, glance at the resulting numbers; if they're not divisible by 8, nudge to a preset or custom size that is.

Custom size only kicks in when use_custom_size is true. If you typed values into custom_size_x/custom_size_y but the output still matches a preset, it's because the toggle is off. Flip use_custom_size to true to make those fields take over.

Pick a resolution your model was trained for. This node makes it easy to select 128×128 or some exotic ratio, but the model still has opinions. SD 1.5 is happiest around 512, SDXL around 1024-equivalent areas - go far outside that and you get duplicated subjects or mush. The preset list includes small sizes for latent-upscale base passes and other tricks, not as final-render suggestions.

It only outputs numbers. There's no image or latent coming out - just width and height. You still need an EmptyLatentImage (or equivalent) downstream to turn those numbers into something the sampler can use.