Dimensions by Ratio

Most dimension nodes make you think in pixels. This one lets you think in shape. PureRatio (displayed as "Dimensions by Ratio") takes an aspect ratio and hands you back a width and height sized for whatever model family you're running - no calculator, no "wait, what's 7:9 in pixels again?"

It's the node to reach for when you're sketching a workflow and you know the look you want (tall, cinematic, square) but don't care about the exact number. Pick a ratio, tell it whether you're on SDXL or SD 1.5, and it outputs two INTs you can drop straight into EmptyLatentImage or any other node's width/height input.

How it works

Under the hood it's one multiplication. Each ratio has a built-in base scale for each model family - SDXL's is anchored at 1024, SD 1.5's at 512, because that's the pixel footprint those models were trained on. The node multiplies your ratio by that base times adjust_scale, then rounds up to integers. So 9:16 on SDXL gives you 810×1440, and 1:1 on SDXL gives you exactly 1024×1024. Same ratio, SD 1.5, and the numbers come out at the 512-ish scale that model actually likes.

That model base is the part to understand, because going off a model's native resolution is how you get the stretched limbs and doubled heads the community is always complaining about. SDXL was trained on a set of discrete ratios around 1024px; SD 1.5 wants roughly 512px. This node bakes that in so you don't have to remember it.

The inputs that matter

Only four, and you'll mostly touch two:

• ratio - nine options from 1:1 through 9:21 and 13:19. This is the shape you want.
• adjust_scale - a float from 0 to 10, default 1.0. Scale the whole thing up or down. At 1.0 you get the model's native footprint; 2.0 doubles it, 0.5 halves it.
• model - SDXL or SD 1.5. Choose whichever checkpoint you're actually sampling with.
• order - default (width,height) or swapped (height,width). The swap just flips the two outputs so you can go portrait without rewiring your graph.

Outputs are width and height, both INT. Wire them wherever a resolution is expected.

The honest caveats

Two things will trip you up. First, and this is the ironic one for a pack whose whole point is Wan video: there's no Wan option in the model list. PureRatio only knows SDXL and SD 1.5 bases. If you're building a Wan pipeline, use the dedicated wan Dimensions or wan Dimensions 720 presets in this pack instead - they encode the actual resolutions Wan 2.x was trained at. PureRatio's SDXL numbers are great for stills and for the image you feed into an I2V workflow, but don't pretend it's a Wan resolution node.

Second, the outputs aren't guaranteed to be multiples of 8 - a common requirement for the VAE latents most diffusion models (Wan especially) run on. 810×1440 from that 9:16 example isn't 8-aligned. If your sampler complains or you see odd seams, round the result up with a "nearest multiple of 8" math node before it hits the latent.

Install

Same story as every node in this pack - and it's a good one, because there's nothing to install beyond the node itself. No requirements.txt, no model downloads, no GPU-busting dependency.

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

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

then fully restart ComfyUI (not just refresh the page).

One gotcha: the README inside the repo is inherited from the original pack and still tells you to clone corbin-hayden13/ComfyUI-Better-Dimensions.git. That installs the original, which has no Wan nodes and no PureRatio in this form. Use the URL above, not the README's.

Verdict

For brainstorming aspect ratios before you commit to a resolution, it's the fastest tool in this pack. For anything where an exact, known-good resolution matters - which is most of the time once you're actually generating - the preset nodes are the safer pick.