Wan 2.2 Size Preset

Text-to-video on Wan 2.2 starts from nothing - there's no reference image to tell you what resolution to use. You just need a width and height that the model won't choke on, and the rules are strict: divisible by 8, aspect ratio between 1:3 and 3:1, and ideally a size the model actually trained on. Manually calculating that every time you want to test a new aspect ratio is the kind of busywork that makes people swear at their own workflow.

Wan 2.2 Size Preset is the t2v sibling of the pack's aspect-ratio helper, and it's simpler in the best way: no image input, no computation, just two dropdowns and three outputs. Pick an aspect ratio and a size tier, get back a resolution that's legal for Wan 2.2, wire the integers into your latent.

How it works

Where the Aspect Ratio Helper derives the ratio from an input image, this node derives it from a dropdown. There are 15 predefined ratios covering the whole supported range - portrait entries from 1:3 Ultra-Tall Portrait through 9:16 and 3:4, 1:1 Square, and landscape from 6:5 and 4:3 up through 16:9 Widescreen, 21:9 Ultra-Wide and 3:1 Ultra-Wide Landscape.

For the chosen ratio it runs the same "smart hybrid" sizing as the helper: it consults the baked-in table of known Wan 2.2 training sizes first, and only calculates a fresh size when your ratio doesn't match a documented one. The size_preset dropdown - tiny through gigantic, roughly 200K to 2M pixels - sets the pixel budget. All generated dimensions are divisible by 8.

One genuinely useful design decision: the node computes and shows all six size options for your ratio, but only outputs the one you've selected. So you can iterate between "fast preview at tiny" and "final render at large" by flipping one dropdown, with no rewiring - the info_text output shows you the whole ladder so you know what you'd get at each tier.

Outputs

• width and height - integers straight into your Empty Latent / Wan size node.
• info_text - a string listing all six size options for the ratio, plus which one you selected.

Installing it

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

Restart ComfyUI, or install "AAA Metadata System" via ComfyUI Manager. No model downloads. The pack is CC BY-NC for non-commercial use; commercial use needs a paid license from the author.

Common issues

• "There's no 16:9 button that matches my monitor" - 19.5:9 and 21:9 are in there, plus everything between. If you truly need an oddball ratio, the Aspect Ratio Helper (with an image) is the better tool.
• Renders look soft at tiny - that's the point of the tier; ~200K pixels is for fast iteration, not final output. Plan to jump to medium or large for the real run.
• Generated dimensions surprise you - the "smart hybrid" prefers known training sizes over exact ratio math, which is what makes Wan behave. info_text shows all the options so nothing is a surprise.

It's a small node that does one thing, and that's fine - in a Wan t2v workflow, where a wrong resolution costs you a long, slow generation run, "one dropdown for ratio, one for size" is exactly the kind of guardrail worth having. The fact that it knows Wan 2.2's actual training sizes puts it ahead of the generic aspect-ratio calculators.