Skip to content
A CRT terminal on a checkered table, dithered with Blue Noise in 1-bitThe same terminal before dithering

13Ordered

Blue Noise Dithering

Blue noise dithering compares each pixel with a threshold mask whose values are spread as evenly as possible, with no repeating structure for the eye to pick out. The idea comes from Robert Ulichney’s work on digital halftoning, and the mask is built with his void-and-cluster method from 1993. It gives a fine, natural grain, with the speed and frame-to-frame stability of ordered dithering.

Author
Robert Ulichney
Year
1993
Complexity
high

How Blue Noise dithering works

PixelSynth tiles a 64×64 threshold mask, 4,096 ranks, across the image. Each pixel turns white when its brightness exceeds the threshold at its position, exactly as with Bayer, and nothing is carried between pixels. What differs is the mask: its ranks are arranged so that at every gray level the lit pixels are spread evenly, with no clumps and no lattice.

“Blue” describes the noise spectrum: its energy sits at high frequencies, the fine detail the eye blends most easily, with almost none at the low frequencies where noise reads as blotches. At 64 pixels, the tile is large enough that its repetition is practically invisible.

Threshold ranks, top-left 8×8 of 64×64
239136321662631291232726652138
2676139840212954188338463477831
31612019344210956996816572919
949558169227552517329820491037
235538371471478365250830814058
189433132898219690124011268234
4467461145397232418533739767
14532506200834531619318025843508
  • A 64×64 matrix of 4096 ranks is tiled across the image; a pixel turns white when its brightness exceeds the threshold at its position.
  • That gives 4097 distinct gray levels, each reproduced by a fixed dot pattern.
  • Showing the top-left 8×8 corner of the full 64×64 matrix.
Pseudocode
M = 64×64 matrix of ranks 0…4095
for each pixel (x, y):
  t = (M[y mod 64][x mod 64] + 0.5) / 4096 * 255
  pixel = gray < t ? black : white
# with a palette, t shifts the color before nearest_color

When to use it (and when not to)

  • Best for: photos and gradients that should look natural, and animation.
  • Works well with: every palette, from 1-bit to Game Boy, PICO-8 or NES.
  • Watch out for: slightly less edge detail than error diffusion, since the mask doesn’t adapt to the image.
  • In animation: Very stable. Each pixel’s threshold depends only on its position, so areas that don’t move stay identical from frame to frame, with no shimmer. The grain stays put, like a film grain.

Blue Noise vs other algorithms

The same terminal in 1-bit, center crop at actual size.

Blue Noise compared with Bayer 8×8, Floyd–Steinberg, Random
AlgorithmMethodAnimationLook
Blue Noise64×64 threshold matrixStableFine even grain, no grid
Bayer 8×88×8 threshold matrixStableVisible cross-hatch
Floyd–SteinbergError to 4 neighbors (100% carried)FlickersSharper, flickers in animation
RandomRandom thresholdStableClumpy white noise

Try Blue Noise on your image

Open the editor with Blue Noise already selected, drop in a photo or an animated GIF, choose the processing scale for the dot size, pick a palette and export the result. Free, no account, no watermark, and your image never leaves your device.

Open Blue Noise in the editor

Blue Noise FAQ

Blue Noise or Floyd–Steinberg?

Floyd–Steinberg keeps a little more detail in stills. Blue Noise is steadier: its dots don’t shimmer in animations, and it has no worm patterns.

What’s the difference between Blue Noise and Void-and-Cluster?

Both masks are built with the void-and-cluster method. Blue Noise uses a large 64×64 mask, so no repetition shows; Void-and-Cluster uses the classic 16×16 array, whose faint period gives a more organic, slightly structured texture.