Scaler | Pixel Deltas

Scaler | Pixel Deltas is the fiddler's scaler: instead of multiplying, it adds and subtracts. Two inputs, width_delta and height_delta (both INT, default 0), and it resizes each side by exactly that many pixels. Add +64 to the width, get a width 64 pixels wider; use -32 on the height, lose 32 pixels. Output is a scaler wire for the pack's Resample nodes and the inpaint Pre node.

It's the least glamorous sizing option here, and honestly it's the one you'll reach for least. But "fiddly" isn't the same as useless - there are two places this node is genuinely the right tool.

When a delta beats a multiplier

First, multiple-of-8 fixing. Diffusion latents want sizes in multiples of 8 (the VAE downsamples in 2× stages), and a pre-existing size is often a multiple of 8 minus a couple of pixels. Instead of recomputing a scale factor, you add the tiny delta that snaps you onto the grid. Set width_delta to whatever the remainder needs and the arithmetic disappears.

Second, adjusting one axis when the other is already right. Say your crop is correct at 1024 wide but you want a touch more headroom vertically. A multiplier scaler would change both sides; Scaler | Pixel Deltas changes exactly the axis you name. It's also handy as a manual "crop to a bit less" control when you're prepping for a compositing or inpainting pass and want to shave a known number of pixels off the canvas.

Using it with the rest of the pack

As with every scaler, it only decides the size - the resampler you wire alongside it decides the method. A small pixel nudge is usually a mild resize, where Resampler | Triangle or Resampler | Mitchell-Netravali are safe picks; a large negative delta (heavy shrink) calls for Resampler | Area to avoid aliasing. And yes, you can go negative - that's the whole point. Set both deltas to 0 and the output is the input size, which is a useful "pass-through, but make it a proper resolution" switch in a workflow.

Installing

Part of comfyui-some-image-processing-stuff:

cd ComfyUI/custom_nodes
git clone https://github.com/wmpmiles/comfyui-some-image-processing-stuff

Restart ComfyUI, or install via ComfyUI Manager (search "Some Image Processing Stuff"). No models, no dependencies beyond PyTorch. Like the other scalers, its wire is required by the Resample node it feeds - leave it unconnected and you get an error, not a default.