DeepStereo: Image Effects

Here's the thing nobody says about Magic Eye images: the texture does more work than the depth map. Your eyes lock onto repeated pattern, and the separation between repeats carries the 3D. If that pattern is blurry, low-contrast, or mushy, the whole image won't fuse - it'll just look like static. This node exists to fix that, and for stereogram work its real job is sharpen.

It's the pack's general image-effects box: one effect dropdown with thirteen choices, a strength dial, and a couple of blend knobs.

How it works

The effect enum splits into two families:

• sharpen and blur - plain image filters. strength (0–1) controls how hard they're applied. Sharpen on a noise texture before it hits the stereogram generator is the money move; the pack's own README advises against blurry, overly complex patterns, and high contrast reads better.
• Blend modes - multiply, screen, overlay, darken, lighten, color_dodge, color_burn, hard_light, soft_light, difference, exclusion. These composite the image against a solid blend_color (a hex string, default #808080 - it accepts with or without the #, and 3-digit shorthand too).

Two extras make the blends usable rather than instant-clip-art:

• preserve_brightness (default on) tries to keep overall luminance steady so a multiply doesn't black out the whole pattern.
• strength lets you dial any effect back, which is what keeps a blend mode a subtle tint instead of a full takeover.

Single output: processed_image, which you can loop back into the stereogram pipeline or anything else in the graph.

What it's actually good for

For this pack, two workflows. First, sharpen a RandomNoiseGenerator or ProceduralTextureGenerator texture before it goes into StereogramGenerator - crisper repeats fuse more reliably. Second, use a blend mode like difference or exclusion against a tint to build a patterned texture that's more interesting than flat noise, or to force a color theme across an otherwise random pattern. The blend modes are also perfectly happy operating on depth maps, which is a quick way to combine two maps into one.

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 use ComfyUI Manager → search ComfyUI-DeepStereo). No models, no weights, no first-run downloads - this node is pure PIL math, so it's one of the most forgiving things in the pack to experiment with. You can sharpen, blend, and blur all day and never wait on a download.