ImageSimpleResize

Most resize nodes make you pick a target width and height. ImageSimpleResize instead has you pick a single target size and tell it which edge that size applies to - the node figures out the other dimension for you, preserving aspect ratio unless you specifically ask it not to. For prepping an image for img2img at a standard resolution without doing the aspect-ratio math yourself, it's a genuinely convenient shortcut.

How it works

You feed it an image and a size, then choose an edge mode: largest scales so the longer side hits size, keeping proportions (a portrait photo and a landscape photo at the same size end up different shapes, both capped on their long edge). smallest does the opposite - the shorter side hits size, useful when you want to guarantee a minimum dimension regardless of orientation. width and height pin exactly one of those dimensions to size while scaling the other to match. all is the odd one out - with no separate second target exposed by default, it's the mode most likely to force both dimensions to the same value, which can distort the image if the source isn't already close to square; if you need two different explicit dimensions, that's what the optional size_override is for.

If you connect a vae, the node also encodes the resized image straight into a LATENT for you - one fewer node than resize-then-VAEEncode separately.

The inputs and outputs that matter

• image - required.
• size - the target dimension, 16 to 32768, default 512 (a classic SD1.5-era default).
• edge - largest / smallest / all / width / height, controls what size measures against.
• size_override (optional) - a second target dimension; use it alongside edge: all or wherever you need width and height to differ rather than match.
• vae (optional) - connect it to also get a LATENT output; skip it if you only need the resized IMAGE.

Two outputs: IMAGE, the resized pixels, and LATENT, populated only if you wired a VAE in.

How to install it

• ComfyUI Manager: search "ComfyUI-Chibi-Nodes", install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/chibiace/ComfyUI-Chibi-Nodes/, restart.

No models to download and no extra dependencies beyond what a normal ComfyUI setup already has for image I/O. It appears under Chibi-Nodes/Image after install. The pack overall is a small, personally-maintained set of utility nodes - the README calls them "experimental" and admits version 2 shipped "with more bugs" - worth keeping in mind as a general calibration for how much polish to expect, though a resize operation like this one is about as low-risk as custom node logic gets.

Common issues

Unlike Chibi's own Loader node, whose width/height fields step in multiples of 8 (matching what SD-family latents expect), ImageSimpleResize's size field has no such constraint - you can land on an odd number like 513. In practice this usually isn't fatal: ComfyUI's VAE encode step crops to the nearest compatible size internally rather than erroring, so a slightly odd resize target mostly costs you a pixel or two of cropping rather than breaking the run. But if you're chasing an exact output resolution, round to a multiple of 8 yourself rather than trusting this node to guarantee it.

The other thing to watch is the LATENT output's silent dependency on vae - if downstream expects a populated latent and you forgot to wire a VAE into this node, the failure shows up at the sampler, not here, which can send you looking in the wrong place first.