wan Dimensions 720

wan Dimensions 720 is the 480 list's bigger sibling - literally. Where the 480 tier tops out at a longest side of 832, this node's presets are those same shapes scaled up by about 1.5× to a max of 1280, with the classic 9:16 and 16:9 crops added. Same idea, more resolution, which in Wan-land means more quality and more VRAM.

If you're generating Wan 2.2 at the tier the community actually considers the quality default - 1280×720 and friends - this is the node that hands you the exact numbers. The 480 tier is for iteration and low-VRAM; this one is for the renders you plan to keep.

How it works

Same mechanism as wan Dimensions: nine hardcoded preset strings, split into width/height, optional swap. The difference is the list, and the list is worth reading closely:

• 720 x 1280 - vertical long, the 9:16 phone-style crop
• 768 x 768 - square, the 512×512 of the 480 tier scaled up
• 744 x 960 / 960 x 744 - short portrait / short landscape
• 1280 x 720 - horizontal long, the standard 16:9 output
• 744 x 992 / 992 x 744 - the mid crops
• 720 x 870 / 870 x 720 - near-square options

Every one of these is a resolution Wan 2.x was trained around at the 720p tier, scaled from the original 480/832 set and, for most of them, aligned to multiples of 8 for the VAE. That's the whole value proposition: you stop guessing whether 1234×694 is "close enough" and start from a number the model actually knows.

The inputs and outputs

Two dropdowns, same as the 480 node:

• dimensions - the nine presets above.
• order - default (width,height) or swapped (height,width).

Outputs: width and height, both INT. Wire them into your Wan pipeline's resolution inputs - the native ComfyUI Wan nodes, Kijai's WanVideoWrapper, or an EmptyLatentImage feeding the sampler.

Why you'd pick this over the 480 node

Honest tradeoff: quality versus time and VRAM. The community's own numbers show 720p Wan generation is where the detail lives - a 14B model at 1280×720 is the standard "showcase" resolution, and the best Wan 2.2 clips in the wild are basically all this tier or higher. But it's also slow: think 20–40 minutes per clip on a 4090-class card at full quality, and 24GB VRAM to do it comfortably. If you have the card, use this node. If you're on 12GB or just iterating on prompts, the 480 tier is the smarter first stop - generate there, upscale the keepers.

One tiny wart: 870 (in the 720×870 / 870×720 presets) isn't a multiple of 8, despite the source's comment claiming the whole list is 8-aligned. It's a non-issue for most workflows, but if a sampler complains about latent dimensions, round that one up to 872 or use a different preset.

Install

Identical to the rest of the pack, and that's the good news - there's no extra weight here. No dependencies, no requirements.txt, no model downloads; the heavy stuff is the Wan weights you already installed, not this node.

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.

And the standing warning for this whole pack: the repo's README is inherited from the original upstream and tells you to clone corbin-hayden13/ComfyUI-Better-Dimensions.git. That's the old image-only pack - no Wan nodes at all. Use the neo-editer URL above.

Verdict

If you've got the VRAM, this is the node you'll actually live in for Wan 2.2. It's the difference between "let me look up what resolution to use" and "oh right, 1280×720 - done."