Undulating Glitch

ComfyUI Undulating Glitch

A small, dependency-free ComfyUI node pack for aligning and mixing four image/video batches with a travelling, undulating, block-glitched transition field.

The node pack was designed for four similarly framed performance videos: the dominant source cycles A → B → C → D → A while an irregular wave boundary crosses the frame. It can also display all four sources simultaneously as moving bands.

Nodes

Four-Way Quick Align

Uses input A as the reference canvas and provides independent alignment controls for inputs B, C, and D:

• X and Y offset in pixels
• scale
• rotation in degrees
• interpolation quality
• white, black, border, or reflected edge filling

The four aligned batches are returned separately so they can feed directly into Four-Way Undulating Glitch Mixer. Input A is passed through unchanged. Inputs B–D are resized to A's dimensions before their transforms are applied.

For isolated heads on a white background, start with fill_mode: white and interpolation: bicubic. Adjust scale first, then X/Y position, and use rotation only for small eye-line corrections.

Four-Way Undulating Glitch Mixer

The fastest route. Connect four IMAGE batches—typically four outputs from VHS Load Video—and send the result to Video Combine.

Outputs:

• images: final four-way composite
• field_preview: coloured preview showing which source controls each area
• transition_edges: mask around active blend boundaries

The mixer includes three movement modes:

• sequential_wipe: a clean undulating transition front
• liquid_wipe: a continuously reshaping, domain-warped liquid front
• four_way_bands: all four sources moving through the frame as bands

Undulating Glitch Field (4-Way)

Generates the effect separately as a continuous MASK. Use this when you want to inspect, modify, blur, distort, or reuse the field.

Four-Way Field Composite

Applies a generated source field to four IMAGE batches.

Liquid Transition Warp

Adds a localized refractive glitch to the mixer's active transition boundary. Connect both images and transition_edges from Four-Way Undulating Glitch Mixer:

Mixer images ──────────────┐
                           ├→ Liquid Transition Warp → Video Combine
Mixer transition_edges ────┘

The warp briefly expands and compresses pixels around the moving boundary, adds ripples travelling along the edge, and can split the red and blue channels for a stronger optical glitch. Areas away from the transition remain unchanged.

Suggested starting settings:

warp_strength: 12
warp_width: 1.4
ripple_amount: 3
ripple_scale: 3
ripple_speed: 2
rgb_split: 1.5
interpolation: bicubic
color_fringe: off
color_fringe_strength: 0.08
water_ripple: off
water_ripple_strength: 1.5

Use a negative warp_strength to reverse the lens direction. Reduce rgb_split to zero for a purely liquid/refraction effect.

Enable color_fringe for a restrained optical grade on the two slopes of the active wave: opposing cyan/magenta bias, slight desaturation, and a small contrast lift. This is separate from rgb_split, which physically offsets the channels. Start with color_fringe_strength: 0.08; values above about 0.15 become intentionally stylized rather than subtle.

Enable water_ripple to add faint alternating compression and expansion bands around the primary transition, like a short refractive wake travelling with the wave. The spacing and decay are derived automatically from the transition mask. Start with water_ripple_strength: 1.5; the practical subtle range is approximately 0.75–2.5 pixels.

Installation

Git clone

From your ComfyUI directory:

cd custom_nodes
git clone https://github.com/jsterlingvids/ComfyUI-UndulatingGlitch.git

Restart ComfyUI, then search for Undulating Glitch.

To update later:

cd ComfyUI/custom_nodes/ComfyUI-UndulatingGlitch
git pull

Manual installation

1. Download the repository ZIP and extract the folder into:

   ComfyUI/custom_nodes/ComfyUI-UndulatingGlitch
   

2. Restart ComfyUI.

3. Search for Undulating Glitch or open:

   image/video/undulating glitch
   

No extra Python packages are required; it only uses PyTorch already included with ComfyUI.

Suggested starting settings

For a liquid-glitch wave travelling left to right through all four sources:

mode: liquid_wipe
cycle_frames: 128
angle: 0
wave_count: 2.4
wave_amplitude: 0.18
wave_speed: 5
noise_amount: 0.035
noise_scale: 3
noise_speed: 2
block_size: 48
block_jitter: 0.018
tear_height: 20
tear_amount: 0.012
glitch_speed: 7
edge_softness: 18
wave_overlap: 0.08
blend_curve: smoothstep

At 24 fps, cycle_frames: 128 makes one complete A → B → C → D → A cycle last about 5.3 seconds. wave_overlap: 0.08 starts the next boundary just before the previous one finishes leaving the frame, keeping the motion nearly constant without a hard reset.

In liquid_wipe, wave_amplitude controls the overall depth of the advancing lobes, wave_count controls their density, and wave_speed controls how actively the large and small ripples crawl along the transition edge. Whole-number wave speeds produce a seamless complete effect cycle.

transition_fraction is used only by sequential_wipe. Liquid mode uses a continuous multi-front timing model instead: wave_overlap: 0.00 places the outgoing and incoming fronts at opposite edges simultaneously, while values around 0.05–0.12 create a small, smooth overlap. Higher values put more than one transition visibly inside the frame.

For harder digital blocks, increase block_jitter, lower edge_softness, and use hard blending. For a flowing liquid boundary, reduce block_jitter and tear_amount, then increase wave_amplitude and edge_softness.

Input preparation

The four videos should ideally share the same:

• frame rate
• dimensions
• duration/frame count
• framing and subject scale

Use Four-Way Quick Align when composition or head placement differs slightly. The mixer's frame_alignment setting can then trim, loop, or stretch shorter batches.

Long videos and RAM-safe streaming

Four 1280×1280 float-image batches can consume tens of gigabytes before compositing. For long clips, add VHS Meta Batch Manager and connect its meta_batch output to:

• all four VHS Load Video nodes
• Four-Way Undulating Glitch Mixer
• Liquid Transition Warp
• every VHS Video Combine output

Start with frames_per_batch: 16 at 1280×1280. The mixer preserves its global frame offset between VHS sub-executions, and the warp uses the same offset for continuous ripple animation, so batching does not reset the wave at every chunk. Video Combine keeps one encoder open and produces one final video.

The alignment and trim-shortest paths also reuse original tensors when no resize or transform is required, avoiding redundant full-batch copies.

Publishing

The repository is already structured like a normal ComfyUI custom node. Before Registry publishing:

1. Create a Publisher and API key on the Comfy Registry.

2. Replace YOUR_COMFY_PUBLISHER_ID in pyproject.toml.

3. Install comfy-cli, then run:

   comfy node publish
   

A release should include a short demo GIF/video and a downloadable example workflow. The demo is the sales pitch; the node name is merely the paperwork.

Status

Prototype 0.8.0. The core effect, continuous liquid timing, transition-warp, water-ripple, color-fringe, RAM-safe VHS meta batching, tensor handling, and quick alignment transforms are implemented and covered by tests, but the pack still benefits from real-world testing inside your specific ComfyUI installation and VHS workflow.