Compression that stays literally invisible to the human eye.
ATTOZIP uses the XYB color space, which models how human eyes respond to brightness and color. Because AttoZip compresses after converting the image to XYB, it removes only the color differences that the eye naturally cannot detect, allowing much stronger compression without visible changes.
AttoZip’s encoder measures every change using Butteraugli, a vision-based algorithm that predicts exactly when a human will notice a difference. During compression, the encoder stops reducing data as soon as the predicted error becomes visible, ensuring the final image looks identical to the user.
Instead of applying one uniform quantization like JPEG/WebP, AttoZip uses adaptive quantization per block. Complex areas (hair, texture, grain, edges) automatically get higher quality, while smooth areas get more compression. This selective protection keeps all fine details intact without wasting space.
AttoZip uses Gabor-like filters, variable block sizes up to 256×256, and smart prediction models. These reduce the amount of data needed without introducing typical artifacts such as blockiness, ringing, or smudging. Because the transforms are designed to mimic natural image structures, the output remains visually perfect.
AttoZip handles gradients with multi-resolution transforms that avoid banding, and it preserves micro-contrast by preventing over-smoothing of low-contrast patterns. Other formats blur these subtle transitions, but AttoZip keeps them intact, which is essential for a “visual lossless” appearance.
if the eye can’t see the difference, then the data can be removed.
This is fundamentally different from JPEG/WebP/AVIF, which focus on pixel accuracy or bitrate targets. AttoZip specifically targets perceptual invisibility, which is why it can compress aggressively while staying visually identical.