A mask is a black and white image for instructing "whether to target / exclude" a part of the image for processing.
White: Process (Apply)
Black: Do not process (Protect)
Gray: Adjust intensity (* Depending on the function)
Mask in AI
In image generation AI, it is mainly used in Inpainting to redraw only a part of the image.
AI regenerates only the white part, and leaves the black part as the original image.
Note:
Normal Inpainting cannot handle gray (translucent) masks, so it is arbitrarily processed as binary of white or black.
If you want to do something like "change lightly" using a mask with gradation, you need a special technique called Differential Diffusion.
What is Output Channel?
Normal images express colors with three channels: R (Red), G (Green), B (Blue), but images with transparent backgrounds (such as PNG) contain an additional channel called A (Alpha).
This is the information governing transparency.
Handling of Transparent Images in ComfyUI
This is a slightly confusing point, but Stable Diffusion itself cannot directly handle transparent images.
Therefore, when you load a transparent PNG image into ComfyUI, it is internally separated into two: "RGB Image" and "Mask".
Let's look at the output of the Load Image node.
IMAGE (RGB)
Since the alpha channel information disappears, the transparent part is filled with black.
MASK
The alpha channel (transparency) information is extracted.
In ComfyUI, the transparent part is expressed as "White".
Beware of differences from general software
In many software, masks are created as "transparent part = black", so people accustomed to image editing software such as Photoshop and Affinity Photo may be confused, but please consider it as a different thing.
It will be easy to understand if you follow the flow: Part to process is white part of mask → Alpha channel is added to white part and becomes transparent image.
Joining and Splitting RGBA
How can I save it with the background transparent again after processing is finished?
Combines IMAGE (RGB image) and MASK to convert it into a single RGBA image (transparent image).
If you connect this to the Save Image node, you can save it as a transparent PNG.
🟪Split Image With Alpha Node
Conversely, it is a node that separates an RGBA image into RGB and MASK.
🚨 Error Trap: Confusion of RGBA and RGB
On the wire of ComfyUI, both RGB image (3 channels) and RGBA image (4 channels) flow as the same IMAGE type. You can't tell by looking.
However, AI (KSampler and VAE) basically accepts only 3-channel RGB images.
Therefore, if you inadvertently connect an RGBA image after Join Image With Alpha to VAE Encode etc., the following error will appear.
RuntimeError: Given groups=1, ... expected input to have 3 channels, but got 4 channels instead
If you get scolded saying "I expected 3 channels but 4 channels came!", please suspect if the image is RGBA somewhere.
Custom nodes for background clipping etc. may output RGBA suddenly.
