Image Composition

This involves overlaying two images or arranging them side by side.

While this is a simple drag-and-drop operation in Photoshop, it can be quite a struggle to achieve in the node-based ComfyUI.

Although there are scenes where it is necessary, such as in Inpainting processing, like color correction, giving up and using a paint tool is also a good choice.

Overlaying Images

This is the operation of simply placing one image on top of another.

ImageCompositeMasked Node

ImageCompositeMasked.json

This is the basic node for overlaying images.

  • destination: The background (base) image.
  • source: The foreground (top) image.
  • x / y: Position adjustment based on the top-left corner.
  • resize_source: If set to true, the source image is stretched to the same size as the destination (Note: it will be distorted if the aspect ratio is different).

Centering the Image

ImageCompositeMasked-Align_center.json

It is possible to center the image by performing coordinate calculations with nodes like the familiar position: absolute; top: 50%; left: 50%; in CSS, but... well, honestly, it's a hassle (；・∀・)

Overlaying Transparent Images

This is the core of layer composition in ComfyUI.

If you want to overlay an "image with a transparent background", simply connecting the images will not work.

In ComfyUI, you need to rethink this as a process of "replacing the masked part with the source image".

Steps to Composite Transparent PNGs

ImageCompositeMasked-with_Alpha.json

We use the same ImageCompositeMasked node, but we use the mask input.

    1. When you load a transparent image, you get a "mask of the transparent part" from the MASK output.
    1. Connect this mask to the mask input of the ImageCompositeMasked node.
    1. Then, only the masked part is replaced by the source image (e.g., pink sky).

Handling Distortion Due to Size Differences

If resize_source is set to true, the source image is forcibly stretched to the same size as the destination image.

ImageCompositeMasked-with_crop-pad.json

This is not a concern if the background is a solid color or pattern, but it will be distorted in the case of photos. The easiest solution is to pad/crop the two images to the same size in advance.

  • 🟪 Pad the foreground image to the size of the background image.
  • 🟨 Crop the background image to the size of the foreground image.

Practical Example: Combining with Segmentation

ImageCompositeMasked_Segmentation.json

The same logic applies when you want to "change the color of only the dress part".

    1. Create a "dress mask" from the input image using segmentation.
    1. Connect EmptyImage (solid color image) etc. to source.
    1. Set resize_source to true (It's OK because it doesn't matter if a solid color is distorted).
    1. Now only the dress part is filled with the specified color.

It may be hard to grasp at first, but you will get used to it gradually.

Stitching Images (Concatenation)

Combine images side by side or vertically. This is useful when you want to create comparison images.

Image Stitch Node

Image_Stitch.json

A simple stitching node.

🪢 Image Concatenate From Batch Node

Image_Concatenate_From_Batch.json

You can arrange 3 or 4 images by using multiple Stitch nodes, but if you want to arrange batch (multiple set) images in a grid, this node handles it smartly.

Mixing Images (Blending)

Image Blend Node

Image_Blend.json

This is like the "blend mode" of layers in paint software. Standard nodes can only do very simple blending, but you can adjust the strength of the composition with blend_factor.

  • blend_factor: The closer to 1.0, the stronger the source image becomes.

EmptyImage Node

EmptyImage.json

A node that just creates a solid color image.

How to Set Color

Setting the color is tricky; instead of the familiar hexadecimal or RGB, you set it in 24-bit color (decimal).

The calculation formula is as follows:

Decimal Value = (Red × 65536) + (Green × 256) + Blue

Example: If you want to create pink (RGB: 255, 192, 203)

255 × 65536 + 192 × 256 + 203 = 16,761,035

Enter the value calculated in this way into the color parameter.