S1Z35 - SDXL

There are two ways to set your resolution in ComfyUI: type numbers into EmptyLatentImage every time, or hand the job to a dropdown. S1Z35 - SDXL is the second way, and its one real trick is that it ships with the exact list of aspect ratios SDXL was actually trained on. If you've ever squinted at a workflow and wondered where "1152x896" keeps coming from, this is the node that answers.

What it is and why it exists

It's a size-preset node: pick a preset, get a width and a height. That's the whole job, and it's a legitimate one, because SDXL is a 1024×1024-native model that was multi-aspect trained at specific buckets - 1152×896 (4:3), 1216×832 (3:2), 1344×768 (16:9), 1536×640 (21:9) and their rotations. Generate way off those trained ratios and you get stretched anatomy and weird crops; generate at them and the model does what it knows how to do. This node's sdxl.csv ships with exactly those eight buckets, which is a nicer default than most "resolution" nodes bother with.

SDXL is no longer where the ecosystem's energy goes - Flux took the SFW photorealism crowd in 2024 and Z-Image took the "SDXL successor" crown in 2026 - but the SDXL fine-tune and LoRA libraries are enormous, and people still run it daily. If you're one of them, a preset node saves you from typing the same five numbers into every workflow you rebuild.

How it works

The engine is shared across all eight nodes in the ComfyUI-S1Z35 pack. At creation time it reads presets/sdxl.csv, parses each width, height, label line into a dropdown entry like Square (1:1) (1024x1024), and when you pick one it pulls the numbers back out and hands them to you. No model files, no GPU work, no API. The label is just a name - the (WxH) suffix is where the real values live.

Inputs and outputs that matter

There are only two inputs:

• preset - the dropdown. This is what you actually set.
• swap_dimensions - a boolean, off by default. Flip it to get the portrait version of a landscape bucket without editing anything. Square (1024x1024) stays square; Landscape (3:2) (1216x832) becomes 832×1216.

Outputs are width (INT) and height (INT). Wire them straight into EmptyLatentImage's width and height - that's the classic use - or into anything else that takes an integer size, like a conditioning size or an upscale node.

Installing it

The pack installs like any other, and this one is unusually light - there's no requirements.txt, nothing to download, nothing to configure:

cd ComfyUI/custom_nodes
git clone https://github.com/K3nt3L/ComfyUI-S1Z35

Restart ComfyUI and the node shows up under S1Z35/Resolution. Or use ComfyUI Manager → Install Custom Nodes and search for "S1Z35".

Common issues

• Your CSV edits don't show up. The preset list is built when the node is created. Edit presets/sdxl.csv, then delete the node and add a fresh one (or restart ComfyUI) - the old node keeps its stale choices.
• "Missing sdxl.csv (512x512)" in the dropdown. The fallback when the file is gone or empty. Outputs will be 512×512, so don't be surprised if your latent is tiny.
• Malformed lines are silently skipped. Keep the format width, height, label and the CSV parser ignores anything it can't read - which is handy, but also means a typo just quietly drops a preset.
• It's not a resizer. This node outputs numbers. If you feed an oddball size to SDXL and the result looks wrong, that's the model complaining about an off-training ratio - pick a bucket closer to native instead.

The pack's author keeps the presets editable precisely so you can tune them. For SDXL you mostly won't need to: the defaults already match what the model was trained on.