wan Dimensions

This is the node this pack exists for. When the fork's author took the old Better Dimensions code and pointed it at video, the whole point was that Wan has its own set of trained resolutions, and nobody had bothered to hardcode them for you. wan Dimensions (the 480 list) is that list, ready to drop into a Wan pipeline.

Wan 2.1 and 2.2 - Alibaba's open video models, the ones the local generation scene settled on - were trained at a specific set of resolutions, roughly a 480p tier where the longest side tops out at 832px, and a 720p tier scaled up from there. Pick a resolution Wan actually likes and your clips behave; pick something arbitrary and you're fighting the VAE from the first frame. This node exists so you don't have to remember which side is 480 and which is 832.

How it works

Mechanically it's the simplest node in the pack: nine hardcoded preset strings, split into width and height, with an optional swap. No math, no model files, nothing to go wrong.

The nine presets are the interesting part. Square 512 x 512 sits in the middle for quick test generations on the small Wan 1.3B and 5B models. Around it are the portrait and landscape crops - 480 x 832 for a tall shot, 832 x 480 for a wide one, plus in-between options like 496 x 640 and the near-square 480 x 580. Every one of them is a resolution Wan was actually trained around, which is exactly why you'd reach for this instead of typing numbers by hand.

The inputs and outputs

Two inputs, both dropdowns:

• dimensions - the nine presets above.
• order - default (width,height) or swapped (height,width), to flip portrait/landscape without rewiring.

Outputs: width and height, both INT. Wire them into the width/height inputs of whatever ComfyUI Wan setup you're running - the native Wan loaders, Kijai's wrapper, or a plain EmptyLatentImage before the sampler.

When to use the 480 tier

Wan generation is slow and VRAM-hungry; the 480 tier is the pressure-relief valve. A 14B model at 832×480 runs far faster and leaner than its 720p sibling, and the standard production pipeline in the community is exactly this: generate at 480p, then upscale the finished clip. Quality-first power users on 24GB cards jump straight to the 720 tier, but for iteration, testing prompts, and low-VRAM cards, this list is the sensible home.

Two honest notes

First, a couple of presets - 660 and 580 as dimension values - are not multiples of 8, even though the rest of the list is. If your Wan sampler complains about latent dimensions, nudge to the nearest 8. Second, this node is a dropdown, not a solver: if you need an arbitrary resolution snapped to a ratio, that's Better Image Dimensions_v2's job. This one is for "give me a known-good Wan number, now."

Install

Light as it gets - no dependencies, no model downloads, one small Python file. If you're generating Wan at all you already have the heavy stuff (the Wan weights); this node adds nothing to that.

Via ComfyUI Manager: search ComfyUI-Better-Dimensions-wan-v2, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/neo-editer/ComfyUI-Better-Dimensions-wan-v2

then restart ComfyUI fully.

Watch for the stale README: it's inherited from the original upstream pack and tells you to clone corbin-hayden13/ComfyUI-Better-Dimensions.git - which gets you the old image-only pack with none of the Wan nodes. Clone the neo-editer URL above.

Verdict

Small, dumb, and exactly what a Wan beginner needs. Install the pack, and this node quietly stops a whole class of "why does my video look wrong" from ever happening.