Quantize
Reduces the image to N total colors using k-means clustering for whole-image palette reduction.
Quantize uses k-means clustering to find the best representative colors for the whole image, then remaps every pixel to its nearest match, for a flat look like an indexed-color image. Unlike Posterize, which snaps each channel to evenly spaced levels, it treats each RGB color as a whole, so the palette it picks follows the image.
Wire a palette into the Palette input to snap every pixel to the nearest color in that palette instead (a perceptual match, the same one Dither uses); Colors no longer applies. With nothing wired, k-means picks the colors.
The result is deterministic: the same image and Colors value always reduce to the same colors, so a batch of near-identical frames does not shimmer from frame to frame, and a document looks the same after reopening as when it was saved.
Parameters
Section titled “Parameters”Pins
Color
Total number of unique colors in the output. Uses k-means clustering to find the best representative colors for the entire image. Lower = more stylized, higher = more faithful
K-means restart attempts. Higher = better color selection but slower. The best result across all attempts is kept
Apply ordered (Bayer 4x4) dithering to reduce visible banding between quantized colors.
Creates a stippled pattern that simulates smoother gradients with fewer colors.
Auto-enabled when a Palette is connected (drives the color set)
Usage Tips
Section titled “Usage Tips”- 8 to 16 colors gives a retro, pixel-art feel.
- Feed the Palette input from a Palette, Palette From Image, or Palette From Gradient node to remap to an exact set of colors.
- Quantize’s own Dither is an ordered Bayer pattern. For error diffusion (Floyd-Steinberg, Atkinson) against the same palette, use the Dither node.