DeepStereo: Random Noise Generator

This is the texture workhorse of the pack. Where ProceduralTextureGenerator builds a pattern from your source image, this node is a straight procedural pattern generator with twelve noise types - and it's tuned for the one constraint that actually matters in stereograms: the pattern has to tile seamlessly and repeat at roughly the width your eyes will lock onto.

The twelve noise_type options: rgb (random color noise), grayscale, colored_dots, perlin, voronoi, kaleidoscope, waves, cellular, fractal, spiral, interference, crystalline. The README's favorites for Magic Eye work are kaleidoscope, crystalline, and interference - the symmetric types tend to fuse more reliably than pure random static.

How it works

Three required inputs define the canvas and the pattern period:

• output_width / output_height - the final image size (64–4096).
• pattern_width (default 100) - the width of the repeating pattern. This is the one that matters. The README says match it to your max_separation in the stereogram node, and that's not decoration: if the pattern repeats wider than the separation range, your eyes have nothing consistent to lock onto.

Then the per-type controls, most of which you only touch for specific noise types:

• noise_scale - scale of the noise pattern.
• frequency (0.1–50) - pattern frequency for the wave/interference types.
• octaves (1–8) - noise layers, for fractal.
• symmetry (2–16) - symmetry points for kaleidoscope.
• color_variation - how much colors wander.
• dot_density and base_color - for colored_dots only.
• seamless_tiling (default on) - makes the pattern tile edge-to-edge without a seam. Leave it on; a visible seam is a fast way to kill fusion.
• random_seed - 0 means a fresh random pattern every run; set it for reproducibility.

One output: noise_texture, wired into StereogramGenerator's texture input.

Getting it to work

Start with pattern_width equal to your stereogram's max_separation and let the defaults carry you. If the result won't fuse, check three things: is seamless_tiling on, is the pattern repeating at roughly the separation width, and is the pattern high-contrast enough to track? The README's texture tips apply here too - high contrast, nothing blurry, pattern width consistent with separation. If you like a look, pin random_seed so it doesn't regenerate every queue run.

Installing it

Standard pack install:

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

Restart ComfyUI, or search ComfyUI-DeepStereo in ComfyUI Manager. No model downloads, no first-run waits - it's numpy and PIL doing the generating, so you can burn through all twelve noise types in an afternoon.