S1Z35 - Qwen

Qwen-Image is Alibaba's 20B image-generation flagship, and this is the pack's node for it. It's also the only node in the whole S1Z35 pack that shows up in Google at all - a handful of impressions, enough to prove someone out there is trying to figure out how to size a Qwen-Image workflow. Fair. It's a size-preset node: pick a preset, get width and height, wire them in.

What it is

A resolution-preset node from the ComfyUI-S1Z35 pack, built for the Qwen-Image family. The qwen.csv ships with four presets: 1328×1328 square, 1472×1104 (4:3), 1584×1056 (3:2), and 1664×928 (16:9). Fewer options than the SDXL or Flux 1 nodes, but they're all in Qwen-Image's real working range, and the aspect ratios are the ones you'll actually want.

A word on what you're sizing for, because it matters. Qwen-Image is a 20B MMDiT - the model behind the best-in-class glyph rendering in the family, especially for Chinese text and posters. It's heavier than the models most people run daily, so it needs real VRAM, and its whole identity is text-heavy, composition-complex prompting rather than tag soup. The node doesn't do any of that work - it just makes sure your latent is a size the model is happy with before you send it into the sampler.

How it works

Same shared engine as every node in this pack. At node creation it reads presets/qwen.csv, parses each width, height, label line into a dropdown entry, and on selection extracts the numbers back out. No model loading, no dependencies beyond Python's stdlib, no GPU work.

Inputs and outputs that matter

• preset - the dropdown with the four Qwen sizes. The only thing you'll set.
• swap_dimensions - boolean, off by default. Flips width and height, so 4:3 (1472x1104) becomes portrait instantly.

Outputs are width (INT) and height (INT). Wire them into EmptyLatentImage (or any size input) ahead of your sampler. For text-heavy Qwen output you'll often want to generate at one of these native-ish buckets and let upscaling handle any extra resolution - pushing a 20B model past its comfortable size is how you get wobbly glyphs.

Installing it

Light install, like the rest of the pack - no requirements.txt, no model files to fetch, nothing to configure:

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

Restart ComfyUI and it appears under S1Z35/Resolution. Or use ComfyUI Manager → Install Custom Nodes and search for "S1Z35".

Common issues

• Edits don't show up immediately. The preset list is built when the node is created. Edit presets/qwen.csv, delete the node and re-add it (or restart) to see new entries.
• "Missing qwen.csv (512x512)" - the fallback when the file is missing or empty, and it will cheerfully give you a 512×512 latent.
• Four presets feels thin. It's a small list, no argument. If you regularly run Qwen-Image at a size that isn't there - say a tall vertical for a poster - add it to the CSV in width, height, label format. The parser picks it up on the next fresh node.
• Missing glyphs or soft text isn't the size's fault. That's usually the sampler or a too-aggressive resolution. Stick near the shipped sizes and upscale instead of brute-forcing.

It's a small node doing one small job, and for Qwen-Image users that job is "don't feed the 20B model a size it wasn't trained on."