Image Resizing and Cropping

It may seem like a simple task of enlarging an image or cropping it to a square, but it is a very important process in image generation.

  • Model's Optimal Resolution: Models have a "resolution where they perform best".
  • Saving VRAM: If you inadvertently load and process a 4K image, you will instantly run "Out of memory".
  • Unifying Materials: In image compositing, etc., it is necessary to align the sizes of multiple materials.

Since this is a frequently used task, let's firmly understand the differences between each node.

Getting Image Information

Get Image Size Node

Outputs the width, height, and batch size (count) of the image as numerical values.

get_image_size.json

Resize

Resize Image/Mask Node

A node that allows you to switch between several resizing methods.
Basically, this one node covers most of the necessary processing. (By the way, masks can also be resized.)

WorkflowsResize Image/Mask Node
resize_imagemask_scale_by_multiplier.json

scale by multiplier

Scales vertically and horizontally by a "multiplier".

For example, 0.5 makes the vertical and horizontal dimensions half, and 2.0 makes them double.

resize_imagemask_scale_dimensions.json

scale dimensions

Forcibly changes the resolution to the specified width and height.

  • crop
    • disabled : Distorts the image if the aspect ratio is different.
    • center : Maintains the center and crops (discards) the protruding parts instead of distorting.
resize_imagemask_scale_longer_dimension.json

scale longer/shorter dimension

Resizes the image specifying only the long side (longer) or short side (shorter), maintaining the aspect ratio.

resize_imagemask_scale_width.json

scale width/height

Resizes the image specifying only either the width or height, maintaining the aspect ratio.

resize_imagemask_scale_total_pixels.json

scale total pixels

Resizes the image to match the specified total number of pixels while maintaining 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 match the size of a reference image.

Previously, it was necessary to get the size of the reference image and pass it to another node, but now this combines it into one.

  • match: Connect the image you want to use as a reference.
  • crop
    • disabled : Distorts the image if the aspect ratio is different.
    • center: Maintains the center and crops (discards) the protruding parts.
resize_imagemask_scale_to_multiple.json

scale to multiple

Resizes the vertical and horizontal dimensions to be multiples of N.

Details are explained in Why can I only generate resolutions that are multiples of 8?, but in short, diffusion models cannot handle resolutions that are not specific multiples due to VAE limitations.

Basically, ComfyUI often adjusts automatically internally, so there isn't much need for this node unless you encounter errors with non-specific resolutions, or when you want to "perfectly match" the pixels of the input and output.

ImageScaleToMaxDimension Node

Resizes the image so that the long side becomes the set size, maintaining the aspect ratio.
(Example: Whether it is a portrait or landscape image, make the longer side 1024px)

imagescaletomaxdimension.json

Padding

Padding is the process of adjusting the size by adding margins (black bars, etc.) around the image.
Depending on the node, this margin part can be output as a mask, so it is used as preparation for Outpainting.

ResizeAndPadImage Node

Resizes to the specified resolution and fills the missing parts with padding.

resizeandpadimage.json

Since this node cannot output the padding part as a mask, there are practically no occasions to use it.

Pad Image for Outpainting Node

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

pad_image_for_outpainting.json

The margin part is output as a mask.

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

Crop and Other Edit Operations

ImageCrop Node

Crops a part of the image in a rectangle by specifying x, y coordinates and width/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 is a node included in kijai/ComfyUI-KJNodes.

It is like a node that combines the above resize, crop, and padding into one.

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)

Previously, this was used frequently to resize to multiples of N, but now that core nodes handle this, it is not used much.

A Little Application

Let's try some slightly complex image processing by combining the nodes so far.

Crop Image to Half

crop_to_half_size.json
  • Get the size of the image
  • Calculate half the width with Simple math node
  • Input the calculated width to ImageCrop node and crop to half