Mask On Canvas

Same idea as its sibling Image On Canvas, but for a MASK - and the important difference is that this node outputs an IMAGE, not a mask. So it's really about seeing masks, previewing them, and building colored overlays, not about passing a mask tensor along.

Why bother? Because inpainting is only as good as your mask, and most mask-producing nodes are bad at showing you what they produced. White in a ComfyUI mask means "this region will be re-denoised" - that's the convention every inpainting node relies on - so before you burn a sampling pass you want to actually look at what's covered. This node turns a mask into a visual you can preview or save: tint it red like a Photoshop quick-mask over a colored backdrop, or render it white-on-black as a clean mask plate. That's also handy for compositing masks into a fixed-size canvas when you're building regional or area-conditioned workflows that expect a specific canvas layout.

How it works

The MASK tensor is converted to grayscale (it auto-detects whether values are 0–1 or 0–255, a nice touch), then a canvas is built at your container_width × container_height and filled with background_color - default black, because white-on-black is the natural mask look. The mask is resized with LANCZOS and pasted at your x/y, clamped and cropped to the canvas bounds.

The mask_color field is the interesting knob. Give it a color (default #FFFFFF) and the mask is drawn in that color with the grayscale mask acting as its alpha channel: bright pixels - mask value 1.0, the region that changes - paint opaque, dark pixels let the background show through. Leave it empty and you get plain grayscale pasted opaquely.

The inputs that matter

• mask - the MASK tensor to draw.
• background_color - hex, default #000000 (black, the natural backdrop for a mask plate).
• container_width / container_height - canvas size, default 512×512.
• mask_x / mask_y, mask_width / mask_height - position and rendered size, defaults 0,0 and 256×256.
• mask_color - hex tint, default #FFFFFF. Clear this field to get plain grayscale.

Output: IMAGE.

Where people get burned

• The default mask_color is #FFFFFF, which is not empty. Leave it alone and you get a white-tinted overlay, not grayscale. To get the grayscale mode you have to actively clear the field to an empty string - easy to miss, and the two modes look genuinely different.
• Grayscale vs. colored are not the same paste. Colored mode alpha-composites, so only the bright parts of the mask paint and your canvas color shows through the shadows. Grayscale mode pastes the whole rectangle opaque - the black parts of the mask cover the canvas color too. If your mask has a black background, grayscale mode will eat your background color.
• Remember the convention: white = the region that changes. A white-on-black plate reads as "this is what gets edited." If it looks backwards, it's the mask semantics, not the node.
• Only the first mask of a batch is drawn. Feed single masks, or split the batch.
• The output is IMAGE, not MASK. It's for preview, compositing, and annotation. If your downstream inpaint node needs a MASK tensor, keep the original mask wire and feed it directly - don't try to pipe this node's output into a mask input.

Installation

Same tiny pack as Andalusi Image Loader and Image On Canvas. No requirements.txt, no model downloads, nothing beyond ComfyUI's stock dependencies.

cd ComfyUI/custom_nodes
git clone https://github.com/AndalusiLLC/comfyui_andalusi.git

Restart ComfyUI, or use ComfyUI Manager and search for "Andalusi." It's a small personal pack by Kornesh Kanan (Kobaisy) - clean and readable, but effectively unmaintained, so don't hold your breath for updates.

Troubleshooting

• Mask looks reversed - white is "will change," black is "stays." If your tint fills the unmasked area, your mask upstream is inverted.
• Background color is invisible - you're in grayscale mode with a mask that has a black background, which covers the canvas. Give it a mask_color or convert the mask to a white-on-black form first.
• Can't wire the output into an inpaint mask input - expected. This outputs IMAGE; keep the original MASK wire for that.