AZ Tools

Image Dither (Floyd-Steinberg, Atkinson, Bayer)

Image

Classic 1-bit dithering in your browser. Pick a photo or screenshot and apply one of seven algorithms: pure threshold (cut at any luminance), Floyd-Steinberg error diffusion (the standard for natural images), Atkinson (the classic Apple Mac / Macintosh / HyperCard look), Sierra Lite (a faster diffusion variant), or ordered Bayer matrices in 2×2, 4×4, or 8×8 sizes (best for repeating textures and authentic Game Boy / e-ink display aesthetics). Pick custom foreground and background colors — defaults are black and white, but you can use any pair (e.g. amber+black for Apple II look, or Game Boy green tones). Optional pixel scale (1×–6×) preserves the dot pattern at larger sizes without smoothing.

How to use

  1. Pick an image with the Choose file button or drag and drop it.
  2. Choose an algorithm: Floyd-Steinberg for photos, Atkinson for that classic Macintosh look, Bayer for an authentic retro game / e-ink texture.
  3. Adjust the luminance threshold (50% / 128 is neutral) and optionally swap foreground/background colors or invert.
  4. Scale up (×2 / ×4) without smoothing to keep the dot pattern crisp at larger sizes.
  5. Click Download to save the result as PNG.

Frequently asked questions

Which algorithm should I pick?
Floyd-Steinberg is the safe default for photographs — it diffuses error to neighboring pixels and produces a smooth, organic look. Atkinson is the Macintosh classic — it only spreads ¾ of the error, giving a brighter, slightly stippled appearance that's recognizable from Susan Kare-era Mac graphics. Bayer 2/4/8 matrices give a tileable, regular pattern that looks like retro game console graphics or e-ink panels.
Does this work on color images?
Yes. The image is first converted to luminance (per ITU-R BT.601: 0.299·R + 0.587·G + 0.114·B), then dithered to two colors. To keep color, look at the Image Posterize or Image Channel Splitter tools.
Why does the output look pixelated when scaled up?
That's intentional. Image scaling uses nearest-neighbor (no smoothing) so the dot pattern stays crisp instead of becoming a smudged grayscale. This matches how pixel art and dithered displays are normally upscaled.
Is the dithering deterministic?
Yes — for a given image, algorithm, threshold and colors, the result is always identical. None of the algorithms here use randomness; the Bayer ones use a fixed matrix and the error-diffusion ones are deterministic raster-order scans.

Related tools