Image Resizing and Cropping

Making an image bigger, cropping it to a square... these are simple tasks, but they matter a lot in image generation.

  • The model's ideal resolution: Every model has a resolution where it performs best.
  • Saving VRAM: Accidentally load and process a 4K image, and you'll hit "Out of memory" in no time.
  • Matching your inputs: For things like compositing, you need several images to be the same size.

You'll do this a lot, so it's worth getting a solid grasp of how each node differs.

Getting Image Information

Get Image Size Node

Outputs the image's width, height, and batch size (number of images) as numbers.

get_image_size.json

Resize

Resize Image/Mask Node

A node that lets you switch between several resize methods.
This one node covers most of what you'll need. (It can resize masks too, by the way.)

WorkflowsResize Image/Mask Node
resize_imagemask_scale_by_multiplier.json

scale by multiplier

Scales width and height by a multiplier.

For example, 0.5 halves both, and 2.0 doubles both.

resize_imagemask_scale_dimensions.json

scale dimensions

Forces the image to the width and height you specify.

  • crop
    • disabled : Distorts the image if the aspect ratio is different.
    • center : Keeps the center and crops off (discards) whatever sticks out.
resize_imagemask_scale_longer_dimension.json

scale longer/shorter dimension

Resizes while keeping the aspect ratio, based only on the long side (longer) or short side (shorter).

resize_imagemask_scale_width.json

scale width/height

Resizes while keeping the aspect ratio, based only on the width or the height.

resize_imagemask_scale_total_pixels.json

scale total pixels

Resizes to the specified total pixel count while keeping the aspect ratio.

Calculated as 1024 * 1024 = 1.00MP.

Target Size Total Pixels Setting Value
512 × 512 262,144 0.25
768 × 768 589,824 0.56
1024 × 1024 1,048,576 1.00
1536 × 1536 2,359,296 2.25
resize_imagemask_match_size.json

match size

Resizes the image to the same size as a reference image.

You used to have to get the reference image's size and pass it to another node, but now it all happens in one.

  • match: Connect the image you want to use as a reference.
  • crop
    • disabled : Distorts the image if the aspect ratio is different.
    • center: Keeps the center and crops off (discards) whatever sticks out.
resize_imagemask_scale_to_multiple.json

scale to multiple

Resizes so that width and height are multiples of N.

I go into detail in Why can I only generate resolutions that are multiples of 8?, but because of how the VAE works, diffusion models can't handle resolutions that aren't a certain multiple as-is.

Some node along the way often adjusts this for you, but this comes in handy when you get errors unless the resolution is exactly right, or when you want the input and output pixels to match exactly.

ImageScaleToMaxDimension Node

Resizes so that the image's long side matches the size you set, keeping the aspect ratio.
(For example, the longer side becomes 1024px whether the image is portrait or landscape.)

imagescaletomaxdimension.json

Padding

Padding adjusts the size by adding margins (such as black bars) around the image.
Some nodes can output the margin as a mask, which makes padding a common first step for Outpainting.

ResizeAndPadImage Node

Resizes to the specified resolution and fills the rest with padding.

resizeandpadimage.json

This node can't output the padding as a mask, so you'll hardly ever use it.

Pad Image for Outpainting Node

Adds the specified number of pixels of margin to the top, bottom, left, and right of the image.

pad_image_for_outpainting.json

The margin is output as a mask.

  • feathering: Blurs the boundary between the margin and the image. Affects only the mask.

Cropping and Other Edits

ImageCrop Node

Cuts out a rectangular part of the image, given x/y coordinates and a width and height.

imagecrop.json

ImageRotate Node

Rotates the image by 90 / 180 / 270 degrees.

imagerotate.json

ImageFlip Node

Flips the image horizontally / vertically.

imageflip.json

Resize Image v2 Node

This node comes from kijai/ComfyUI-KJNodes.

It's basically the resize, crop, and padding above rolled into one node.

Parameter Name Description
width / height Target width and height (no change if 0)
upscale_method Interpolation method during resizing (nearest, bilinear, etc.)
keep_proportion stretch, pad, crop, etc.
pad_color Color for padding (RGB)
crop_position center, top, bottom, left, right
divisible_by Resizes to a resolution that is a multiple of this value (e.g., 32, 64)

I used to use this a lot for resizing to multiples of N, but core nodes can do that now, so I don't use it much anymore.

Putting It Together

Let's combine the nodes so far to do something a little more involved.

Crop Image to Half

crop_to_half_size.json
  • Get the size of the image
  • Calculate half the width with the Simple math node
  • Feed that width into the ImageCrop node to crop the image in half