Animation_RGB_Mask

Ever seen those animations where different parts of an image change one after another - the background blinks first, then the mid-ground, then the subject - and wondered how the workflow knows which region is "active" at any moment? Animation_RGB_Mask is that machinery. It generates an animation of color-coded masks: each frame assigns distinct colors to distinct regions, and the regions sweep, rotate, or pulse over time according to one of 12 modes. Downstream, you interpret each color as a region id and drive per-region effects.

It's the direct sibling of Animated Shape Mask in the Eden pack. Where the shape node gives you clean grayscale sweeps, this one hands you labeled regions - think of it as a paint-by-numbers video where the palette and the regions both animate.

How it works

Internally it builds an Animation object at your width/height with num_colors colors and total_frames frames, then renders each frame and stacks them into an IMAGE batch. The modes:

• concentric_circles/rectangles_ - nested rings or boxes expanding inwards or outwards; each ring gets its own color.
• rotating_segments_ - a color wheel's worth of wedge segments spinning clockwise or counter-clockwise.
• pushing_segments_ - like rotating but the segments visibly push the previous one along the frame.
• vertical/horizontal_stripes_ - stripe columns or rows marching left/right or up/down.

The two controls you'll actually tune:

• num_colors (default 3) - how many distinct regions exist per frame. More colors = more regions, so more independent animation channels downstream.
• bands_visible_per_frame (default 1) - how many color bands are visible at once; raise it to overlap regions.
• angle (default 0) - rotates the whole pattern, which is how you angle stripes or wedges to match your subject.

Outputs are IMAGE (the stacked animation), plus num_colors, width, and height echoed as INTs.

What it's actually for

The practical pattern, which Eden's own video workflows use, is: color-code the frame, then map each distinct color to a different prompt or mask target. Region 0 gets inpainted with prompt A, region 1 with prompt B, and because the regions are color-labeled you can extract each one as a clean mask per frame. Combined with the KB's inpainting insight - that masked inpainting uniquely leaves everything else pixel-identical - you get frame-consistent regional animations without full-video regeneration. That's the whole trick behind those "objects morph in sequence" videos: it's not one prompt animating the scene, it's a labeled mask deciding where each change is allowed.

Installing

Part of edenartlab/eden_comfy_pipelines (Eden.art nodesuite), installed once for everything:

cd ComfyUI/custom_nodes
git clone https://github.com/edenartlab/eden_comfy_pipelines.git
cd eden_comfy_pipelines
pip install -r requirements.txt

Or ComfyUI Manager → search "Eden". No extra model downloads - it's pure numpy/pytorch rendering.

Gotchas

Watch your resolution: total_frames at, say, 1024×1024 stacked into one tensor gets heavy fast. Also, like its sibling, the output is IMAGE not MASK - if you're feeding a KSampler, convert to MASK and threshold per color first. And the echo outputs (num_colors, width, height) are handy for keeping downstream nodes in sync, but they're plain values - don't expect them to update live as you drag sliders mid-run; you'll re-run the graph like any ComfyUI node.