DOGMA ResolutionK v1.0.11

Some nodes are infrastructure, and pretending otherwise wastes your afternoon. This is one of those: you give it a K, it gives you back a long side in pixels. resolution_k 5 → 5120, 6 → 6144, 4 → 4096. It's min(8, max(1, k)) * 1024 and nothing else.

Why it's in the pack at all

DOGMA workflows are written in "K" - the studio's phase prompts and notes talk about a 4K phase, a 6K phase, a 12K max - and every one of those nodes downstream (DOGMA ScaleFactor v1.0.11, the crop and fit helpers) works in pixels. Rather than have eight hardcoded INT widgets scattered across a graph that all have to be nudged in lockstep when someone decides the master should be 6K instead of 4K, you have one K and everything derives from it.

The practical use is exactly that: wire long_side into a fit/resize node's long-side input, and now "the whole pipeline is 5K" is a single integer you change once.

The inputs and outputs

• resolution_k (INT, default 5, range 1–8) - the K value. Step is 1, so you get 1K through 8K in whole-K jumps. There is no in-between; 4.5K is not a thing this node will do for you.
• long_side (INT) - the computed pixel figure, which is what you wire onward.

That's it. No image input, no latent, no model. It's a number generator.

Install

Manager → search DOGMA Nodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/axior/ComfyUI-DOGMA-Nodes

Restart. The pack's only Python dependency is scipy>=1.10, and this particular node imports nothing beyond what ComfyUI already has - it can't fail on a missing dependency, only on being wired to a node that expects a different unit.

Where it bites

The value is a long side, not a width and not a megapixel count. If you feed long_side into a widget that wants width on a portrait image, you get a 5120-wide portrait that your VRAM will remember. Pair it with the dimension helper that respects aspect ratio (DOGMA FitDimensions32 v2 takes a long_side and returns width and height for the actual image), and check the previewed numbers before you queue a 12K run on a 12GB card. The 32-pixel alignment those helpers do matters too - SDXL- and VAE-era architectures need dimensions divisible by 8, and the 32 multiples give you margin for the crop/pad machinery elsewhere in the pack.