Empty RGB image

The star of this pack, Pre CFG gradient scaling, can steer a generation toward the colors and layout of any image you feed it - but you don't always have an image handy, and sometimes all you want is "make the whole render lean blue." Empty RGB image is the pack's answer: a tiny generator that produces a solid-color image at whatever size and batch you ask for, ready to drop into gradient scaling as the guide.

It's a one-trick node, but the trick is useful in three spots. First, color influence: feed a plain red image into gradient scaling's input_latent... well, actually into the pipeline that encodes it - the README's own examples show solid-color guides shifting the whole color palette of a result. Second, it's the canvas for experiments: gradient scaling can hardly contradict what the model wants to draw, so a solid color is a soft, controllable nudge rather than a takeover. Third, and this is the part people miss, the grayscale_to_color optional IMAGE input lets you take an existing grayscale image and tint it with your chosen RGB values, turning a luminance-only image into a colored guide in one step.

The inputs are self-explanatory: width and height (default 1024 each, up to 16384), batch_size (default 1), and the r / g / b channels (default 0, i.e. black). Connect grayscale_to_color and the node ignores its own size settings, matching the grayscale image's dimensions instead and applying your RGB tint to it. The single IMAGE output is what you route onward.

It lives in the "image" category rather than the model_patches group, but it's part of the same pack and exists specifically to serve gradient scaling - which is also why the pack ships its companion Gradient RGB image for anything more elaborate than a flat color. Solid color for palette influence, gradients for direction and vignetting.

Installation is the pack standard:

cd ComfyUI/custom_nodes
git clone https://github.com/Extraltodeus/pre_cfg_comfy_nodes_for_ComfyUI

or search pre_cfg_comfy_nodes_for_ComfyUI in ComfyUI Manager and restart. No models, no extra dependencies - it's pure tensor generation.

A blunt but honest tool. It won't impress anyone on its own, but when you want to test "what if the whole image were this color" on gradient scaling, it's the fastest way to find out.