Web Design

Our heroes ship as JPG. WebP would cut them 72%

We measured our own heroes this week instead of quoting a guide. The last three blog heroes ship as JPG at quality 92, sized 2976 by 1632, and weigh 741, 654, and 618 kilobytes. The same three images converted to WebP at quality 80 weigh 210, 184, and 165 kilobytes. Every image came out 72 to 74 percent smaller. One format decision, three pages lighter, zero visible difference at the sizes readers actually see.

Why we shipped JPG in the first place

The choice was boring on purpose. JPG needs no build configuration, decodes on every browser that has ever rendered an image, and the quality-92 setting kept compression artifacts off our heroes at full resolution. Our nightly pipeline generates the hero, converts it to JPG, and copies it into the deploy folder. JPG was the format that added zero steps to that chain, and a pipeline that runs unattended at 11 p.m. prices every added step honestly.

The price of boring turned out to be bytes. Three heroes, roughly two megabytes combined, where the same three images fit in 559 kilobytes as WebP. On the fiber connection this post was written on, nobody notices. On a phone at a train station, that difference is most of a page load.

What the format guides agree on

SpeedVitals' 2026 comparison and Crystallize's breakdown describe the same landscape we measured: WebP lands 25 to 34 percent smaller than JPG at comparable quality in their tests, which matches our 72 percent at the aggressive quality-80 setting once you account for our chosen starting point. Browser support stopped being an argument years ago, as TinyPictures' format guide records: every current browser decodes WebP, Safari included since 2020. ShortPixel's comparison adds the detail that matters for the next step: AVIF typically lands smaller still, at the cost of slower encoding and a thinner tooling story.

The guides disagree on margins, not direction. Every source ranks them the same way: JPG heaviest, WebP in the middle, AVIF smallest. Direction is all a migration decision needs.

The migration that fits in one line

Our pipeline converts heroes with Pillow, the Python imaging library. The entire WebP migration is one extra save call per image, executed at build time on our machine where encoding cost is free. The browser-facing side ships a fallback chain: WebP first, JPG behind it, and every reader gets the newest format their browser speaks. No JavaScript, no runtime conversion service, no new dependency. The images arrive smaller.

The same pattern scales past WebP. When an AVIF encoder lands properly in our toolchain, the chain grows another source entry and the pages shrink again. Each format slot in the chain costs a line, and the browser picks its best automatically.

The catch worth stating

WebP at quality 80 is not pixel-identical to JPG at quality 92 at full resolution. At the resolution readers see, card thumbnails and screen-width heroes, the difference disappeared in our comparison. Archives and print derivatives keep the full-quality JPG masters, and only the browser-facing copies gain the lighter encoding. Keep the master, ship the optimized copy, and the migration costs nothing permanent.

Measure your own three heaviest images this week. Convert them to WebP at quality 80, compare the byte counts, and let your own numbers write the argument. Ours wrote 72 percent.

Sources

The format landscape in this post comes from these guides:

  • SpeedVitals compared WebP and AVIF for 2026, covering compression ratios, encode times, and browser support.
  • Crystallize broke down the same two formats on speed, quality, and support across browsers.
  • TinyPictures published the broader format guide covering AVIF, WebP, JPEG, and PNG with support histories.
  • ShortPixel added the AVIF trade-off detail: smaller files, slower encodes, thinner tooling.

The measurements are ours: three heroes from this blog's November pipeline, converted with Pillow, quality 92 JPG against quality 80 WebP, byte counts read straight from disk.

More in Web Design