Icy ImageScaleBy

IcyImageScaleBy is the ComfyUI-IcyHider wrapper of core ImageScaleBy - same node, renamed class, no functional difference. If you've read the IcyImageScale article, you already know 90% of this one: it's a geometric resize, not a quality upscaler. The one real difference is how you express the target size.

ImageScale asks for an exact width and height. ImageScaleBy asks for a multiplier instead - one scale_by float (0.01–8, default 1) and it scales the image by that factor in both dimensions. scale_by: 2 doubles the size, scale_by: 0.5 halves it. No arithmetic, no checking what the source resolution was, no crop option to worry about - the aspect ratio is preserved by construction.

Inputs

• image (IMAGE)
• upscale_method (enum) - nearest-exact, bilinear, area, bicubic, lanczos, same choices and same advice as its sibling: bicubic or lanczos for general work, nearest-exact for pixel-art/test steps, area for clean downscaling.
• scale_by (FLOAT, 0.01–8, default 1) - the multiplier.

Output: a single IMAGE, sized proportionally.

Why pick this over ImageScale

Convenience, mostly. If your workflow is "always 2x the incoming image" - say you're upscaling a ControlNet condition or prepping a hires pass - the multiplier is more robust than hard-coding dimensions, because it adapts to whatever resolution shows up. It also can't accidentally distort the image the way ImageScale does when you guess the wrong aspect ratio, since there's no fixed target to misjudge.

Same honest caveat as always: this multiplies pixels, it doesn't add detail. For quality upscaling you want a real upscaler (Lanczos/ESRGAN for "more pixels," SUPIR/SeedVR2-style for "more detail" - the KB's upscaling doc lays out exactly which job needs which tool). Reserve this node for resizing jobs.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/icekiub-ai/ComfyUI-IcyHider

Restart, or use ComfyUI Manager (search "IcyHider"). No pip dependencies, no models.

Common issues

• Missing node in shared workflows - install the pack or swap in core ImageScaleBy.
• "The output isn't a multiple of 8." True - this node doesn't snap to latent-friendly dimensions. If the result feeds a VAE/latent path that needs multiples of 8, round your scale_by or follow it with an ImageScale to a clean size.
• "It's softer than I expected." Again, geometric resize. The node is doing exactly what it claims.