AGSoft Image Resize MP

"How big is this image?" is a question people usually answer in megapixels, but resize nodes make you answer in width and height. AGSoft Img Res MP flips that: you say "1.0 MP" and it figures out the exact W×H to hit it while keeping your aspect ratio - then optionally rounds to a model-friendly multiple. It's the laziest way to standardize a bunch of differently-sized images before they hit a sampler or an upscale pass.

It's also the quiet workhorse of the pack's resize family: fewer knobs than the Resize Plus node, but for "make this a standard ~1MP" it's the one with the least friction.

How it works

Give it target_megapixels (default 1.0 - about 1000×1000 equivalent) and the node computes dimensions from the source's aspect ratio, then applies the other two settings that matter:

• multiple_of - rounds the result to a multiple of 2/8/16/32/64/128 (default 0 = off). Turn on 64 for SDXL, 8 for SD1.5, and stop fighting sampler divisibility errors.
• fit_mode - what to do when rounding breaks the perfect aspect fit. pad (default) fits the whole image in the target box and fills the rest with padding_color; crop fills the box and trims the overflow. Since rounding usually nudges the aspect slightly, this decides whether you lose pixels or gain a border.

position (default center) places the image within the padded box when fit_mode is pad - center, corners, or edges. interpolation defaults to lanczos, and area is the better pick for pure downscaling.

Inputs and outputs

Inputs: image, target_megapixels, interpolation, multiple_of, fit_mode, padding_color, position. Outputs: the resized image, plus width and height ints. Note there's no mask output here - if you need the mask resized to match, the Image & Mask Resize (or Resize Plus) nodes are the right choice. This one is purely about the image.

Where you'd use it

Three quick cases. Standardizing a batch of photos of wildly different resolutions before a grid or a dataset job. Prepping a source for a latent that expects ~1MP input. And as a cheap downscale stage: a soft 4K photo upscaled from a 0.35MP base - the community trick for sharp sources - is a target you can express directly in MP here.

Installing it

Pack standard: ComfyUI Manager → search comfyui-AGSoft → install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Art-xmaster/comfyui-AGSoft.git

No models, no extra dependencies - pure PIL/torch.

Common issues

The main gotcha is fit_mode defaulting to pad: if your source aspect doesn't match the rounded box, you get a border you didn't ask for - flip to crop when you want zero padding. And remember the pad happens after rounding, so a border may appear even at multiple_of 0 if the megapixel math lands on a non-integer. If your outputs are mysteriously a few pixels bigger than expected, rounding-up is doing its job; check multiple_of before suspecting a bug.