ColorFill

Sometimes you don't want to edit an image, you just want its dimensions. ColorFill is the node for that: you feed it any image and a color string, and it hands you back a brand-new solid-color IMAGE with the same width and height. That's the entire job, and it's weirder to describe than to use.

It's part of the small, opinionated ComfyUI-TeaNodes pack by TeaCrab (real name in the license: Yuan Qing Liu, MIT). It lives in the TeaNodes/Input category, which is the hint - this is an input generator, not an image editor. It doesn't touch the image you give it at all; it just reads height, width off the tensor and makes a fresh one.

When you'd actually reach for it

• A background behind a cutout. You've removed a background with BiRefNet, and you want a clean solid backdrop behind the subject at the exact right size for an img2img pass. Wire the mask/output image into image, pick a color, done - no manually typing 1024×1024 anywhere.
• A canvas for inpainting or img2img. Drop a colored rectangle where the sampler expects a full IMAGE tensor and you need it to match a reference size.
• Batch consistency. Every run gets an identical-size, identical-color base, which keeps the rest of your graph happy when a dimension mismatch would otherwise kill the queue.

What you actually set

Two inputs, one obvious:

• image - any IMAGE. Only its dimensions are used.
• color - a string, default #7f7f7fff (mid-grey, fully opaque). The reliable formats are #rrggbb and #rrggbbaa. With eight hex digits you get an RGBA fill; with six, an RGB one.

Output is a single IMAGE, ready to wire into VAEEncode, an image-to-image input, or a composite/mask operation.

Gotchas

ColorFill's own parser is the stingy one of the pack. It understands hex and comma-separated integers (127,127,127,255), but anything shorter than six hex digits falls into a try/except that prints Something went wrong here and hands you garbage. If your fill comes out wrong, that's almost always the color string, not the node. (The sibling RandomColorFill uses a fancier parser if you need more formats.) Also note the output has no alpha when you give it six-digit hex - you lose the alpha channel, which surprises people once in a while.

Installing it

From ComfyUI Manager, search ComfyUI-TeaNodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/TeaCrab/ComfyUI-TeaNodes

then restart ComfyUI. The pack declares exactly one dependency, regex_spm (the author's own tiny library), and there are no model downloads. Manager will handle the pip install; everything else comes with ComfyUI core.

It's a tiny node with a tiny job. But when a workflow demands "a flat color image sized exactly like that other image," it beats doing it by hand, and it's free.