FL Image Blank

There's no trick to FL_ImageBlank - it makes a solid-color image at whatever width, height, and RGB values you give it. That's genuinely the whole node. But "generate a blank canvas" turns out to be one of those unglamorous utilities you reach for constantly once you're building non-trivial workflows: a background layer for FL_ImageOverlay when boundary_behavior is set to extend the canvas, a base to composite text or shapes onto with something like FL_TextOverlayNode, or just a placeholder image while you're testing a graph and don't want to load a real file yet.

The inputs are exactly what you'd expect. width and height (64–8192, in steps of 64, default 512×512) set the canvas size - the step of 64 lines up with the dimensions most diffusion models actually expect, so you're less likely to end up with an off-grid size that some downstream node resizes on you unexpectedly. red, green, and blue (0–255 each, default 255/255/255, i.e. white) set the fill color directly. There's no alpha input here - the output is a plain RGB tensor, so if you need transparency, pair this with FL_ImageToMask or FL_ApplyMask to add an alpha channel on top rather than expecting FL_ImageBlank to give you one.

One IMAGE output, and that's it.

Where this actually earns its place in a graph is as the "known quantity" input for anything that needs a starting point but doesn't care what's in it. Testing a new latent-noise workflow and don't want to burn time picking a reference photo? Blank canvas. Building a mask-generation chain and need something to attach FL_ImageToMask to before you've wired in a real source? Blank canvas. Setting up FL_ImageOverlay or FL_RegionPNGOverlay and want a plain background to check your alignment and blend-mode settings against before you drop in the real base image? Same answer. It's the node equivalent of a blank sheet of paper - unglamorous, but it removes a dependency from every experiment that doesn't actually need a real image yet.

Installing it is the standard Fill-Nodes routine - search "Fill-Nodes" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI_Fill-Nodes

• then restart ComfyUI. FL_ImageBlank has zero dependencies of its own beyond whatever the base pack already needs; it's pure tensor math, no models, instant to run.

Troubleshooting, such as it is. There's genuinely not much that can go wrong with a node this simple, which is sort of the point. The one thing worth flagging: keep width/height sane relative to whatever you're compositing this into downstream. If you're using it as a canvas for FL_ImageOverlay's extend_canvas boundary behavior, make sure the blank canvas is actually big enough to hold the overlay at its offset, or you'll end up cropping something you meant to keep. And if a downstream node is expecting an alpha channel and silently misbehaving, remember this node doesn't produce one on its own - that's a separate step, not a bug in FL_ImageBlank.