Y7 Image Size (Presets)

Every model family has its own set of "good" resolutions, and most of them aren't the ones you'd guess. SD1.5 likes 512-ish, SDXL 1024-ish, Flux has its own native buckets, Qwen-Image wants multiples of 28, and video models have their own aspect conventions. Y7 Image Size (Presets) bundles all of that into one node: pick a model family, pick a dimension, get the width and height out as clean integers.

It's from the Y7Nodes pack - the kind of "probably only useful to me" utility the author specializes in, but it's one of the more genuinely time-saving ones if you switch between model families a lot.

How it works

Two dropdowns drive everything. preset selects the dimension set - Default, SD1.5, SDXL, Flux.2, Z-Image, Qwen-Image, Video, or Custom*. dimension then shows only the resolutions that belong to that set, so the dropdown stays clean instead of dumping 108 options on you at once. The dimension list updates dynamically when you change the preset.

Outputs are simply Width and Height as INT. Wire those into an Empty Latent Image, a conditioning-size node, or anything else that takes integer dimensions.

When you want something not in the list, select Custom from the dimension dropdown and the custom_w / custom_h inputs take over - both INT with a 128–7680 range and a 16px step, so they snap to sensible values.

The custom preset

The Custom* option is the flexible one: it loads its own list from custom_dimensions.json in the pack's nodes directory. The format is demonstrated in custom_dimensions_example.json - it's just a list of label/value pairs. If the file is missing or malformed, the node silently falls back to the default set rather than erroring, which is a nice piece of defensive design. This is the option you'd use if your favorite checkpoint has a specific set of resolutions you keep reaching for.

Why you'd reach for it

If you only ever run one model, this is overkill - you know your resolution and you'll type it. Where it earns its keep is multi-model work: testing the same prompt across Flux.2, Z-Image, and Qwen-Image means three different resolution regimes, and this node turns "which dimensions does this family use?" from a lookup into a dropdown. It also keeps resolution choices out of your typing muscle memory - no more accidentally feeding an SDXL-sized latent to a Flux pipeline because you typed 1024×1024 on autopilot.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/yushan777/ComfyUI-Y7Nodes
cd ComfyUI-Y7Nodes
pip install -r requirements.txt

Restart ComfyUI, or use ComfyUI Manager → search "Y7Nodes".

Gotchas

One thing worth flagging: the preset names are approximate buckets, not guarantees of a model's native resolution. Flux.2's ideal canvas, Qwen-Image's 28-multiple constraint, and the like are captured as curated lists here, but if a model ships an updated recommendation you'll want the custom file path rather than relying on the built-in list. Also, remember the node outputs raw integers - there's no latent creation built in, so you still wire Width/Height into whatever consumes them. It's a dimension provider, not a magic box.