DataBend | GlitchNodes

DataBend is the kitchen-sink node of GlitchNodes: slice shifting, channel splitting, hue shifts, wave distortion, compression artifacts, and pixel sorting all in one place. If you want one node that makes an image look "glitched" in the broadest possible sense - think the noise-and-shimmer aesthetic from VHS intros and broken video-game captures - this is the fastest route. It's not the deepest tool in the pack, but it's the most convenient.

The mechanism is straightforward glitch-by-slice. It cuts the image into horizontal and/or vertical bands (controlled by slice_min_size, slice_max_size, and slice_variability), then displaces those slices, and layers in color-channel manipulation. On top of that, glitch_types picks the slice behavior - shift, repeat, mirror, noise, or all - and pattern_frequency / chaos_amount decide how many passes and how violent they are. The three distortion sliders (wave_distortion, compression_artifacts, pixel_sorting) are the seasoning you add on top.

The inputs that matter

• slice_direction - horizontal, vertical, or both. Horizontal slices are the classic "broken analog" look; vertical gives you that sideways tear.
• glitch_types - all (the default) is the most chaotic; repeat and mirror give more recognizable datamosh-style echoes.
• channel_shift_mode - random, rgb_split, or hue_shift. rgb_split with rgb_shift_separate on gives you clean chromatic-aberration ghosting; hue_shift runs the hue around the wheel.
• chaos_amount (0–1) - the master intensity for slice displacement. 0.5 is a good starting point.
• color_intensity (0–1) and preserve_bright_areas (0–1) - the second one is a genuinely useful safety valve: raise it to keep highlights from turning into muddy garbage during color shifting.
• seed - required, and it's your reproducibility knob; same seed, same bend.

Output is image (IMAGE) → PreviewImage/SaveImage, and it handles batches. Input is called images in the schema, matching its batch-first personality.

Installing it

Same as every GlitchNodes node - one pack install covers all 19:

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

or ComfyUI Manager → Install Custom Nodes → search "GlitchNodes". Dependencies: numpy, torch, Pillow, opencv-python, scipy. No models.

Gotchas

Because DataBend stacks so many effects, it's easy to turn a portrait into an unreadable mess in one click. The README's framing applies here more than anywhere: "highly experimental and may break or even error." Practical tip: raise pattern_frequency slowly, keep chaos_amount at 0.3–0.5, and use preserve_bright_areas to keep faces readable. And when you land on a look you like, that seed is your friend - it's the difference between "reproducible" and "never seen again."