Rekked WIP | GlitchNodes

Rekked is the pack's port of the Datamosh-js/datamosh repo (credited right in the source), and it's the easiest way to get twelve distinct glitch looks from a single dropdown. blurbobb, fatcat, vaporwave, castles, chimera, gazette, manticore95, schifty, vana, veneneux, void, walter - each one a different flavor of block shifting, channel mixing, noise, or threshold art. If you're new to glitch nodes and want to see what's possible without reading twenty parameter lists, Rekked is the on-ramp.

Under the hood it flattens the image to raw bytes and runs each mode as a distinct byte-level or channel-level transformation, seeded so every mode is reproducible. schifty does random-length block shifts (the classic horizontal datamosh tear), blurbobb drops random pixel blocks at intervals, chimera mixes channels with noise and grain, fatcat multiplies brightness layers for that blown-out color, and void is pure noise/grain for dark textured looks. The modes genuinely diverge - some are subtle, some are brutal - so the fun is just clicking through them.

The inputs that matter

• mode - the only real control. The 12 options are: blurbobb, fatcat, vaporwave, castles, chimera, gazette, manticore95, schifty, vana, veneneux, void, walter.
• seed - required. Rekked is seeded per-mode, and the same seed + mode gives the same output. When you hit a schifty tear you love, the seed is how you keep it.

Input is image (IMAGE), output is image (IMAGE) → PreviewImage/SaveImage. Batches work, though it's worth testing single frames first because byte-level modes can be slow across a long sequence.

Installing it

Part of GlitchNodes:

cd ComfyUI/custom_nodes
git clone https://github.com/pxl-pshr/GlitchNodes
# restart ComfyUI

or ComfyUI Manager → Install Custom Nodes → search "GlitchNodes". Deps: numpy, torch, Pillow, opencv-python, scipy. No model downloads - it's all NumPy byte manipulation.

Gotchas

Rekked is marked WIP, and it shows in two ways: it's not in the pack README's main node list, and the effects vary wildly in intensity - void and manticore95 can obliterate an image while vana barely whispers. Don't judge the node by one mode. Also, because it works on flat byte arrays, mode behavior is resolution-dependent: the same mode at 512px vs 2K will look different, so match your test resolution to your final one. And the standing warning: "highly experimental and may break or even error." When in doubt, schifty - it's the mode that reliably reads as "datamosh" to anyone.