Image Resize Pixaroma

What it does

Applies one of eight resize modes to the wired image and its optional mask, then outputs the result with its final pixel dimensions. You can also wire a width, height, or longest-side value from another node to drive the size automatically.

Resize modes

Off

Passes the image through unchanged.

Max megapixels

Scales down so the image does not exceed a total pixel count. Uses 1 MP = 1024x1024 to stay aligned with standard AI sizes.

Longest side

Scales so the longer edge hits the number you set. Works for portrait or landscape without picking an axis.

Scale by

Multiplies both dimensions by a factor (2.0 doubles, 0.5 halves).

Fit inside

Fits the image inside a W x H box while keeping the whole image visible and the aspect ratio intact.

Crop to fill

Scales and center-crops so the image completely fills a W x H box. Edges may be trimmed.

Match aspect ratio

Resizes to a chosen preset ratio, either cropping or padding.

Pad

Adds a colored border to reach the target size. The padded area becomes white in the mask (the inpaint region).

Wiring a target size

Connect a value to width, height, or longest_side from another node (e.g. Resolution Pixaroma) to drive the size. Wire only width or only height for an aspect-preserving scale. Wire both for an exact box. longest_side takes priority when connected.

Outputs

image

The resized image.

mask

The resized mask. In Pad mode the added border is white. If you leave the mask input empty and the picture already has a see-through background, that transparency comes out here instead of being lost.

width

Final output width in pixels.

height

Final output height in pixels.

longest_side

The longer of the output width and height.

Pictures with a see-through background

If you feed in a picture whose background has been removed, wire the mask output along with the image into Join Image with Alpha, and you get the cut-out back at the new size, ready to save as a transparent PNG.

You do not need to wire anything into the mask input for this: when that input is empty, the picture's own transparency is used, and it is cropped and resized exactly the same way the picture is. If you do wire a mask in, yours is used instead.

The image output itself always stays a normal picture, on purpose. Transparency travels in the mask in ComfyUI, and a picture carrying a see-through channel cannot be fed to a sampler: it stops the run with an error. Keeping the two apart means this node fits anywhere.

The buttons on the node

The mode chips

Off, Max MP, Longest side, Scale by, Fit inside, Crop to fill, Match ratio and Pad. Pick how the image is resized.

Snap

Rounds the final width and height to a multiple of 8, 16, 32 or 64, which most models prefer.

Resample

How the pixels are recalculated when scaling. Auto suits almost everything.

Upscaling

Whether the image is allowed to grow beyond its original size.

Fill and Crop

In Crop to fill: Fill scales to cover then trims, Crop takes a piece at original size with no scaling.

The anchor grid

Which part of the picture to keep when something has to be trimmed.