⭐ Star Qwen Image Ratio

ComfyUI's built-in Empty Latent Image works fine - if you already know the numbers. The problem is the numbers. Qwen-Image was trained at its own set of native resolutions, and they're not the round figures you'd guess: 1:1 is 1328×1328, 16:9 is 1664×928, 4:3 is 1472×1104. You can type those into an Empty Latent Image manually, sure. You just shouldn't have to.

StarQwenImageRatio is the Qwen-native version of that node. Pick a ratio from the dropdown and it produces an empty latent at the resolution Qwen-Image was actually tuned on, plus the width and height as plain integers for everything downstream.

What you set

• ratio - ten options, all with the resolution baked in: 1:1 (1328×1328), 16:9 (1664×928), 9:16 (928×1664), 4:3 (1472×1104), 3:4 (1104×1472), 3:2 (1584×1056), 2:3 (1056×1584), 5:7 (1120×1568), 7:5 (1568×1120), and Free Ratio (custom) which unlocks custom_width/custom_height (16–8192, stepped by 16) for when your aspect isn't in the list.
• batch_size - 1–64, same as any empty latent.

Outputs: latent → your sampler, width/height → any node that needs the dimensions for math (upscalers, cropping, aspect-aware things).

Why this over the built-in node

Honestly? Two reasons. The resolution table is encoded in the dropdown, so you can't misremember whether Qwen's 16:9 was 1664×928 or 1728×960 - it's right there in the label. And the width/height integer outputs are genuinely useful: build an upscale ratio, size a mask to match, or feed a text node that reports dimensions, all without a separate GetImageSize-style node or hardcoding numbers. It's a small convenience, but it's the right kind of small convenience - the kind that removes a class of typo.

Also worth knowing the model context: Qwen-Image is the 20B MMDiT foundation from Alibaba, Apache 2.0, famous for text rendering. It's heavier than the models most people run daily, but if you're using it, running it at its native ratios is one less variable in the "why does this look off" column.

Install

The pack's usual story: ComfyUI_StarBetaNodes is the retired beta repo, and these nodes now ship in ComfyUI_StarNodes. Install via ComfyUI Manager (search Starnodes) or git clone https://github.com/Starnodes2024/ComfyUI_StarNodes into custom_nodes, then restart. No model downloads, no pip deps - it's a latent-size helper. If you only ever generate at 1:1, you could skip it; the day you need a 7:5 canvas without doing math, you'll be glad it's installed.