ResizeInterpolationControl

This one doesn't resize anything. It's a single dropdown - nearest, bilinear, bicubic, area, nearest-exact, or lanczos - that outputs whichever interpolation algorithm you picked, typed so it slots straight into the interpolation input on DemonAlone's own ImageResizeNode (itself a port of cubiq's well-known Image Resize from ComfyUI_essentials). Think of it as a remote control: instead of setting interpolation directly on the resize node, you set it here and wire the result over.

Why bother with the extra node? Same reason you'd pull any setting out into its own widget - so it's controllable from one place if several resize steps need to agree, or so it's exposed as a clean parameter on a workflow you're handing off, where someone running it via the API can flip one dropdown without touching the graph. Given this ships in the same pack as DemonAlone's XYZ-plot toolkit, it's also a plausible fit for sweeping interpolation methods as one axis of a comparison grid - though that's a reasonable guess at intent, not something the README states outright.

What the choices actually mean

• nearest / nearest-exact - no blending, just picks the closest source pixel. Blocky, fast, and the only sane choice for masks or pixel-exact data where you don't want values invented between pixels. nearest-exact is the corrected version - plain nearest has a well-known off-by-one alignment quirk that nearest-exact fixes, so use it unless you have a specific reason not to.
• bilinear - smooth, cheap, the sensible default for most resizing.
• bicubic - sharper than bilinear, a bit more expensive, generally preferred when quality matters more than speed.
• area - averages pixels rather than sampling them, which is specifically good at shrinking images without moiré or aliasing. Reach for this when you're downscaling significantly.
• lanczos - the sharpest of the six, the classic choice for upscaling. It's a pure interpolator, not a generative upscaler - it can't invent detail that wasn't there - but for "I just need more pixels and the source is already sharp," it's a genuinely competitive, current answer, not a legacy option people have moved past.

The inputs and outputs that matter

• interpolation - the entire input, a six-way dropdown.

Output: interpolation, typed as that exact six-value combo rather than a generic string. That typing matters - see below.

How to install it

ComfyUI Manager: search "DemonAlone-nodes-ComfyUI", install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/DemonAlone/DemonAlone-nodes-ComfyUI, then restart. Pure Python, no models, nothing to download - genuinely one of the lightest nodes in the pack.

Common issues & troubleshooting

It looks like it should do something, and it doesn't. This node has no image input and no image output - it only produces a value. If you dropped it into your graph expecting a resized image out the other side, you've grabbed the wrong node; you still need ImageResizeNode (or whatever resize node you're actually running) downstream.

The wire won't connect. Because the output is typed to this exact six-choice combo, it only plugs into inputs expecting that same set. Try wiring it into a different pack's resize node with a different interpolation vocabulary and ComfyUI will refuse the connection outright rather than silently accepting a mismatched value.

Not worth it for a one-off resize. If you're only resizing once and never reusing the choice, just set interpolation directly on the resize node. This node earns its keep when the value needs to live somewhere shared, not for a single isolated resize step.