Image Resize

Image Resize is the boring, useful one: it changes an image's dimensions using ordinary interpolation. No AI, no model, no VRAM - just resampling pixels the way any image editor does. Set a target size (or a scale factor), pick how it interpolates, and out comes the resized image. That's it, and "that's it" is exactly why you'd use it.

When to reach for it - and when not to

Reach for it whenever you need a specific size and don't need to invent detail. Matching two images before a blend or stitch. Fitting an input to a model's expected resolution. Shrinking something for a preview or a control image. Normalizing a batch to one size. For all of that, plain resize is correct, fast, and lossless-feeling.

Do not reach for it expecting a quality upscale. This is the distinction beginners miss, and our upscaling notes hammer it: interpolation can make an image bigger, but it can't add detail that wasn't there - blow a 512px image up to 2048 with bicubic and you get a soft, smeary 2048. Real upscaling uses a trained model (ESRGAN-family via Upscale Image (using Model), or a latent/diffusion upscale) that hallucinates plausible detail. Use Image Resize to hit an exact target dimension; use an upscale model when you want the result to look better, not just larger.

The settings that matter

The two things you'll set: the target (either explicit width and height, or a rescale factor) and the resampling method. The method is the quality/character dial:

• Nearest - no smoothing; keeps hard pixel edges. This is the one you want for pixel art or masks you don't want blurred.
• Bilinear / bicubic - smooth general-purpose downscaling and mild upscaling.
• Lanczos - the sharpest of the interpolations, good for downscaling detailed images.

WAS's resize also supports supersampling for cleaner downscales. The output is the resized image, ready for whatever's next.

Watch for these

• Aspect ratio. If you set width and height independently to a ratio that doesn't match the source, the image stretches. Decide whether you want a squash or a crop-to-fit.
• Upscaling with interpolation looks soft. Expected - see above. That's a job for an upscale model, not this node.
• Masks and pixel art: use nearest, or you'll blur edges you meant to keep crisp.

Installing it

Ships in WAS Node Suite. Install the pack via ComfyUI Manager (search was-node-suite-comfyui, install, restart) or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/WASasquatch/was-node-suite-comfyui

install requirements.txt, restart. No model downloads for resizing.

Troubleshooting

The node itself is dead simple and reliable. The pack is where trouble lives: WAS Node Suite has been unmaintained since December 2023, and the classic symptom is the whole suite reporting "Import Failed" after a ComfyUI update - a dependency clash (opencv usually, since WAS pins an older version). Reinstall requirements.txt into the correct venv (activate it, or use install.bat) and it loads again.