How to use the image dithering tool
- Choose an image: JPG, PNG, WebP, GIF, BMP or AVIF, up to 25 MB and 24 megapixels. SVG is not accepted.
- Pick a method, the number of tones from 2 to 8 and a dot size in pixels. Choose a dark and a light color for the ends of the scale.
- Tick work in linear light if you want overall brightness to follow the light of the original more closely.
- Watch the preview, then save the PNG at the full size of the original.
The methods
Dithering shows a picture with few tones by arranging dots so that the eye blends them. Error diffusion picks the nearest tone for each pixel and passes the rounding error to the pixels that are still to come. Floyd-Steinberg sends 7/16 of the error to the right, 3/16 below left, 5/16 below and 1/16 below right, and it scans back and forth along the rows to avoid streaks. Atkinson sends 1/8 of the error to each of six neighbors and drops the rest, which gives crisper contrast but loses some detail in the lightest and darkest areas.
Ordered dithering compares each pixel with a fixed repeating pattern of thresholds called a Bayer matrix. The 2 x 2, 4 x 4 and 8 x 8 matrices give coarser to finer patterns. It is fast and has no drifting noise, so it suits animation and pixel-art looks. The plain threshold just picks the nearest tone for each pixel.
A smooth gray ramp from 5 to 95 percent has a mean brightness of 0.5 and Floyd-Steinberg keeps it within 0.1 percent (0.5002), while Atkinson gives 0.5054. The tests check the exact tone of every pixel on a small picture against a separate script.
Color, size and brightness
With two tones the picture uses only your dark and light colors. With more tones, the middle ones are blends between them. The dot size shrinks the picture before dithering and shows each dot as a block, which makes the pattern coarser. Transparent areas are shown over white, and the result has no transparency.
Gray values come from the Rec. 709 weights on the stored color values. Working in linear light converts them to a light scale first, which keeps the average light level closer to the original but makes midtones look lighter. For a plain black and white or gray conversion, see convert image to black and white; for other formats and sizes, the image converter and the image compressor; to change the stored resolution, use the image DPI metadata editor.
Assumptions and limits
- The picture is read at its full size and nothing is uploaded.
- Gray uses Rec. 709 weights, with transparent pixels shown over white.
- The saved PNG uses only the colors of the palette you chose, with no transparency.
- Metadata such as the camera and location is not copied to the new file.
Frequently asked questions
Which method looks best?
It depends on the picture and the purpose. Floyd-Steinberg keeps the most detail in photos. Atkinson gives a classic early-computer look with strong contrast. Ordered patterns are best for flat areas and for a deliberate pattern look.
Why does the preview look different from the saved file?
The preview is a reduced view so the page stays fast. Error diffusion and patterns depend on the pixel grid, so a smaller preview shows a coarser version of the pattern. The saved file uses the full size.
What does linear light do?
Pictures store brightness on a curve. Dithering in linear light spreads dots according to real light, which is more physically correct, but the middle of the picture then looks lighter than most people expect. Try both.
Can I use it for a laser engraver or a thermal printer?
Many people do. Use two tones, a dot size that matches your device's smallest dot, and the PNG output. Test on scrap first.