Nodes/ComfyUI-IcyHider/Icy ImageScale
ComfyUI Node

Icy ImageScale

Resize to an exact size, the honest way

By icekiub-ai·Created 9 months ago·Updated 9 months ago· 31
Icy ImageScale
  • image
  • IMAGE
upscale_method
width512
height512
crop

IcyImageScale is the ComfyUI-IcyHider wrapper of core ImageScale - same node, renamed class so the privacy extension can hide it. And a quick reality check before the details: this is a geometric resize, not a super-resolution upscaler. The KB's upscaling doc splits "make it bigger" into three very different jobs, and this node only does the first one - more pixels, no invented detail. If you want a model to hallucinate plausible new detail (faces, textures), you want an ESRGAN-family upscaler or a tile-ControlNet pass, not this. But for getting an image to a specific resolution, this is the node.

Inputs that matter

  • image (IMAGE) - whatever you're resizing.
  • upscale_method (enum) - nearest-exact, bilinear, area, bicubic, lanczos. For most work bicubic or lanczos (sharper, slightly slower) is right; nearest-exact gives you chunky pixelated scaling (useful for testing and for step-edges like pixel art); area is good for downscaling without shimmer. Don't overthink it - bicubic is the safe default.
  • width / height (INT, 0–16384, default 512) - the exact output dimensions.
  • crop (enum, disabled / center) - what to do when your source's aspect ratio doesn't match the target. disabled squishes/stretches the image to fit exactly; center scales to cover the target and crops the middle. If you're resizing into a fixed-size latent canvas (multiples of 8, remember), center crop is usually what you want to avoid distortion.

Output: a single IMAGE at the requested size.

Where it fits in a workflow

  • Prepping condition images for ControlNet. ControlNets want their condition map at the model's generation resolution - resize your edge map or pose render to match the latent canvas.
  • Standardizing inputs. ImageBatch needs matching dimensions; a quick ImageScale on both sources makes them fit.
  • img2img / hires prep. Feed a scaled image into a sampler's latent path (via VAE encode) to control the exact working resolution.

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 ImageScale.
  • "It stretched my image." You left crop on disabled with a mismatched aspect ratio. Switch to center.
  • "The upscale is soft." Again: this is a resize, not a detail-generator. For real quality upscaling, reach for the actual upscaler nodes - the KB's hierarchy puts pixel upscalers (Lanczos, ESRGAN models) and generative restorers (SUPIR, SeedVR2) in separate jobs, and this node is firmly in the "more pixels" bucket.
CategoryIcyHider Comfy Core

Inputs (5)

NameTypeDefaultDescription
imageIMAGE
upscale_methodCOMBO5 options: nearest-exact, bilinear, area, bicubic, lanczos
widthINT5120–16384
heightINT5120–16384
cropCOMBO2 options: disabled, center

Outputs (1)

NameTypeDescription
IMAGEIMAGE