Push a clip into the palette of an old computer or handheld console. Every pixel is snapped to the nearest color in the palette, and the rounding error is spread into its neighbours, so gradients survive as patterns of dots instead of hard bands.
Palettes
Two-color black and white gives the stark 1-bit look of early Macintosh graphics. Game Boy maps everything onto four greens. PICO-8, Commodore 64, and DB16 each hold 16 colors, enough to keep skin tones and skies recognisable.
Fewer colors mean more visible dots. The pattern carries the detail the palette cannot, so faces in a 2-color dither stay readable at 720p and above but fall apart in a small export.
Error diffusion on video
The dither uses Floyd–Steinberg diffusion in linear light. It is ordered by position, which makes it sensitive to change: a pixel that shifts slightly can reorder the dots for the rest of its row and below. On camera footage the pattern crawls from frame to frame. Many people use that shimmer as part of the look.
For a steadier result, lower Diffusion. The dots thin out and more of the frame settles into flat palette areas, which change only when the picture does.
Output size
Dithered frames are full of high-frequency detail, which is the hardest thing for a video encoder to compress. At the same bitrate the export shows more compression smearing than the source did. Export at the source size rather than scaling down, since shrinking a player window blurs the dots into gray anyway.