Scheduled Binary Comparison | Akatz

Most "binary threshold" nodes give you one threshold for the whole image. AK_ScheduledBinaryComparison gives you one threshold per frame, pulled from a schedule list - which makes it the gate between a float schedule and an animation. You hand it an image batch and a list of thresholds, and it binarizes every frame against its own value. Same idea as the audio-reactive pipelines in the KB's animation write-ups: a schedule becomes an on/off mask, frame by frame.

How it works

For each frame i in the batch, it takes comparison_schedule[i] as the threshold and builds an output of pure 1.0/0.0 pixels. The subtle part is the comparison mode:

• With use_epsilon on (the default), a pixel becomes white when it's equal to or within epsilon[i] of the threshold - |pixel - threshold| <= epsilon, an exact-equality OR'd in. That's a band around the threshold, not a "greater than" test.
• With use_epsilon off, a pixel becomes white when it's >= threshold. That's the plain thresholding most people picture when they hear "binary comparison."

That default matters. If you wire this up expecting a simple high-pass and get a band, flip use_epsilon off. Schedules shorter than the batch get padded by repeating the last value; longer ones get truncated. Both comparison_schedule and epsilon_schedule are expected per-frame lists, the same LIST format the pack's AK_KeyframeScheduler outputs.

Inputs and outputs

• images (IMAGE) - the batch to binarize.
• comparison_schedule (LIST) - the per-frame thresholds. Feed it from AK_KeyframeScheduler, AK_AudioFramesyncSchedule, or any LIST output.
• epsilon_schedule (LIST, optional, default [0.1]) - per-frame epsilon for the band mode.
• use_epsilon (BOOLEAN, default true) - the mode switch described above.
• images (IMAGE) - output, same shape, values pinned to 1.0/0.0.

Installing the pack

Part of Akatz Custom Nodes (akatz-ai/ComfyUI-AKatz-Nodes). ComfyUI Manager: search "Akatz" and install, or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/akatz-ai/ComfyUI-AKatz-Nodes

Restart ComfyUI; it's under 💜Akatz Nodes. Only torch is needed for this node (already in ComfyUI); the pack's other requirements.txt entries (numpy, opencv-python, pydub) belong to its image/audio nodes. No models. Docs live in the author's Notion page and custom GPT, not the README.

Gotchas

• The default epsilon mode is probably not the behavior you guessed. Test with use_epsilon off first if you want straightforward >= thresholding.
• Your schedule must line up with the batch length. Padding repeats the last value, which is fine for a hold but silently wrong if you expected per-frame values that aren't there.
• The output images are hard 0/1, so downstream nodes that expect soft or alpha values will get harsh edges. That's the point - just know it going in.