Resize Multiband

Resize Multiband is the pack's answer to the mismatch problem that keeps biting everywhere else in it. Remember how Compose Multiband and Concat Multiband hard-refuse to join channels with different spatial sizes? This is the node that fixes one side before you hit that wall: it resizes every channel of a MULTIBAND_IMAGE at once, keeping the stack intact.

You could do this with stock nodes - decompose to masks, resize each, recompose - but that's a lot of wiring for something that's one node here. Multiband Resize mirrors ComfyUI's built-in Upscale Image node so closely you'll feel at home: same upscale methods, same width/height/crop combo, same semantics.

How it works

It takes the (B, C, H, W) tensor and runs every channel through ComfyUI's own common_upscale function, so the interpolation quality matches what you'd get from a normal upscale node. The upscale_method dropdown offers the standard set - nearest-exact, bilinear, area, bicubic, lanczos - and crop is either disabled or center. Same choices, same look, applied uniformly across all channels so the stack stays aligned.

The width/height pair has the useful ComfyUI convention: set one to 0 and it's computed from the other while preserving aspect ratio. Both zero means no resize at all (a passthrough), which is a handy way to keep the node in the graph without changing anything.

The inputs

• multiband - required, the stack to scale.
• upscale_method - nearest-exact, bilinear, area, bicubic, or lanczos. For masks and other binary-ish data, nearest-exact keeps hard edges crisp; for smooth feature maps, bicubic or lanczos looks better.
• width / height - target dimensions, 0 meaning "derive from the other."
• crop - disabled or center.

Output is a multiband at the new size, channel names and metadata intact.

Installing the pack

From ComfyUI-Multiband. ComfyUI Manager → search "ComfyUI-Multiband" → Install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/PozzettiAndrea/ComfyUI-Multiband.git
pip install -r ComfyUI-Multiband/requirements.txt

Deps: numpy, torch, tifffile. It leans on core ComfyUI's comfy.utils, which is already there - no extra installs, no models.

The one real decision

Method choice for masks. If you resize a hard segmentation mask with bilinear, you get fuzzy anti-aliased edges that some downstream tooling will reinterpret as soft alpha. nearest-exact keeps the 0/1 structure intact. For everything else - spectral bands, feature maps, depth - the smoother interpolators are usually what you want. And if you resize before composing, you can stop worrying about the mismatch errors entirely: bring the smaller stack up to the larger one's resolution with one node instead of debugging a compose failure.