Which dithering algorithm to choose for laser engraving (Floyd-Steinberg, Atkinson, Stucki...)

Almost nobody explains this in plain terms applied to laser work, so most "settings" tables online never mention it — but the dithering algorithm you use changes the result as much as power or material do. Two identical photos, dithered with different algorithms, can look like two completely different quality levels of engraving.

What dithering is, in short

The laser only has two states per point: burn or don't burn. To simulate a photo's greys, the dithering algorithm decides which points to burn and which to leave blank, so that from a distance the eye blends those points into an intermediate tone. The difference between algorithms is how they distribute those points.

There are two broad families:

• Error diffusion (Floyd-Steinberg, Atkinson, Stucki, Jarvis, Burkes, Sierra): each pixel's rounding "error" spreads to neighboring pixels. Results look organic, with irregular texture similar to film grain.
• Ordered / matrix (Bayer): uses a fixed, repeating pattern. Gives a regular texture, similar to offset print screening — more predictable, faster to compute, and better on reflective materials where an irregular pattern looks messy.

The algorithms, one by one

Floyd-Steinberg — a reasonable starting point for almost everything. Balanced error diffusion, good detail, good compute speed. If you don't know which to use, start here.

Atkinson — diffuses less error than Floyd-Steinberg (only 3/4 of the total, the rest is discarded). The result is smoother with more pure whites, which favors portraits and faces: less noise on skin, more definition around eyes and mouth. It's the algorithm the original 1980s Mac used for its black-and-white screen, and it works well in engraving for the same reason.

Stucki — longer-range error diffusion (spreads to more neighboring pixels). Gives more sharpness and contrast than Floyd-Steinberg, at the cost of a bit more noise in flat-tone areas. Good for images with fine detail: architecture, textured landscapes, illustrations.

Jarvis-JJN — similar to Stucki but with an even wider diffusion pattern. Produces very smooth gradients, ideal for photos with long tonal transitions (skies, soft-lit portraits), but it's the slowest to compute.

Burkes — a simplified version of Stucki, faster, with a result between Floyd-Steinberg and Stucki.

Sierra — similar to Jarvis but faster to compute with nearly identical results. A good option if you like Jarvis but want less processing time.

Bayer (ordered) — a regular pattern, not organic. On paper or screen it looks "less natural" than error diffusion, but on anodized aluminum and other marked metals it avoids the irregular blotching that error diffusion sometimes leaves on very reflective surfaces. It's also the fastest of the group by a wide margin.

Ostromoukhov — an adaptive error-diffusion variant that adjusts its coefficients based on local image tone. Gives better midtone detail than classic Floyd-Steinberg, at a somewhat higher compute cost.

Quick reference: what to use for each case

Situation	Recommended algorithm	Why
Not sure / general use	Floyd-Steinberg	Safe balance of detail and speed
Portrait / photo of a person	Atkinson	Less skin noise, cleaner facial features
Landscape or highly detailed image	Stucki	More sharpness and definition
Photo with soft gradients (sky, diffuse light)	Jarvis-JJN	Cleaner tonal transitions
Anodized aluminum / reflective metal	Bayer	Avoids irregular blotching on polished surfaces
Logo or image with text	Bayer or Floyd-Steinberg	The regular pattern keeps text edges cleaner
Need to process many images fast	Bayer or Burkes	The fastest to compute

How to decide without guessing

The table is a starting point, not a fixed rule — material, the original photo's contrast, and personal taste all change which one looks best in any given case. The way to avoid guessing is comparing several combinations on the same image before engraving the good material.

MEZ Laser includes all ten algorithms described here (plus material-calibrated presets) and a test mode that compares up to 12 algorithm/DPI combinations on the same image before you spend any material.

Once you've picked an algorithm, the other factor that most changes the result is output resolution — covered in what DPI to use for laser engraving.