mmuuai选择器

The name is honest this time. MmuuaiResolutionSelector (shown as "mmuuai选择器" in the node menu) is a selector, not a processor - it never touches a pixel, never resizes anything, never loads a model. You pick a preset and it hands you numbers: width, height, fps, a pick-one-of-twenty value, and a boolean, ready to wire into whatever node downstream wants dimensions. If you've been doing the mental math every time you build an Empty Latent Image for a 9:16 clip, this is the node that does it for you.

Why that's worth a node at all: every model has a native resolution and a divisibility rule it's happy with - SDXL's trained aspect ratios, Flux wanting multiples of 64, video models that care about the short edge. Generate off-ratio and you get stretched bodies and doubled heads, the community's classic complaint. This node keeps your width and height locked to the ratio you asked for and rounds them to a multiple models actually tolerate.

How the math works

It's one small function with three modes, picked by 类型 (type):

• 图片 (image): the 清晰度 preset is treated as the long edge - 1K → 1024, 2K → 2048, 4K → 4096, 8K → 8192 - and the result is rounded to a multiple of 16.
• 视频 (video): the preset is the short edge - 480p → 480 up to 4320p → 4320 - rounded to an even number. On standard 16:9 that gives you the textbook sizes: 854×480, 1280×720, 1920×1080, 2560×1440, 3840×2160, 7680×4320. Flip to 9:16 and you get 720×1280 instead.
• 自定义 (custom): ignores presets and uses 自定义宽度/自定义高度 (custom width/height, 2–16384), rounded to even.

画面比例 (aspect ratio) offers 13 presets from 1:1 through 21:9 and 1:2; 帧率 (fps) is a pick from 16/24/30/50/60. A small JS widget draws the computed resolution right inside the node, so you see the final size before you generate instead of squinting at a downstream latent node.

The inputs that actually matter

• 类型 and 清晰度 - the two you'll touch 95% of the time.
• 画面比例 - set once per workflow and leave it.
• 选择输入 (select input) plus the 输入1–输入20 (Input 1–20) sockets - a mini switch: connect a few integer wires, dial the selector to the one you want, and the value output emits it. Handy for flipping between two sizes or seed values without rewiring.
• 布尔开关 (boolean switch) - a plain on/off that goes to the bool output.

What you get out

Five outputs: width (宽), height (高), fps (帧率), value (选择结果), and bool (布尔值). The classic wiring is width/height into an Empty Latent Image or a video node's size inputs; value feeds a seed or a count; bool gates something further down the graph.

Install

Same as the rest of the pack - ComfyUI Manager (search "mmuuai-resolution-selector") or:

cd ComfyUI/custom_nodes
git clone https://github.com/mmuuai/mmuuai-resolution-selector

Then restart ComfyUI. No dependencies (the pack's pyproject lists none), no model files, no API keys. It's pure local arithmetic.

Two things that trip people up

It computes but never applies - if you want your actual images scaled to that resolution, that's the pack's other node (mmuuai全能节点), not this one. And the UI is entirely Chinese, field names included; if you don't read it, you'll be matching labels by position. The pack is also from a small Chinese author with basically zero community footprint - this node is clean (nothing here phones home), but it's a reminder that any unknown custom node is code you're choosing to run on your machine.