Ordnary

How Nyquist picks the right compression for every image

Nov 16, 2026 1 min read

Instead of one setting for every photo, Nyquist measures each image and keeps the smallest version that still looks the same.

Stijn Jakobs Co-Founder & Chief Technology Officer

Most websites compress every image with the same setting. That setting is too strict for a sky and too loose for small text, so some files are far bigger than they need to be and others look worse than they should. Nyquist™ makes one decision per image instead.

Measure, search, keep

For every image, Nyquist encodes it, measures how close the result looks to the original with a perceptual metric, SSIMULACRA 2, and keeps the smallest version that passes. A short search per format, AVIF, WebP and JPEG, is enough.

What it saves

We measured it on 120 photos at every quality step from 50 to 100, against the best single setting that keeps 95% of them visually identical. JPEG files came out 25% smaller on average, AVIF 13% and WebP 11%.

We also tested whether a trained model could do better by choosing a quality for every region of an image. It couldn't: the encoders' own tuning already does that.

What's next

The research is done. The next step is building Nyquist into R2, so images in public buckets are optimized the first time they're requested. There's no release date yet.

Choose which cookies Ordnary may use. You can change this at any time via Cookie settings at the bottom of every page.