Case study
Mabel Rivero
mabelrivero.com
Every image in the library re-encoded to AVIF, with WebP behind it and the untouched original behind that. No redesign, no rebuild, and nothing deleted.
less for a visitor to download, measured across 19 pages of the live site.
The heaviest page
Averages hide the page that actually
hurts. On mabelrivero.com it is /es/inicio/, which went from
3.0 MB to 1.3 MB —
56.5% lighter. That is the page a visitor on a phone was waiting for.
Page by page
Every page, weighed as a browser loads it
HTML, stylesheets, scripts and images together — not a lab score and not a
projection. Re-runnable at any time, because the page carries both halves of its own
comparison: the AVIF a modern browser takes, and the original still sitting in the
<img> fallback inside it.
| Page | Before | Now | Lighter |
|---|---|---|---|
| Inicio | 3.0 MB | 1.3 MB | 56% |
| Hauptseite | 2.7 MB | 1.3 MB | 53% |
| Home page | 2.7 MB | 1.3 MB | 53% |
| Choreography | 2.4 MB | 1.4 MB | 41% |
| Bio | 1.6 MB | 1.1 MB | 29% |
| Tango holidays tuscany casale de pasquinelli italy | 1.6 MB | 978 KB | 40% |
| Tango holidays crete greece with michael domke | 1.5 MB | 1.0 MB | 35% |
| Tango feast autumn devon england with fernando guidi | 1.4 MB | 941 KB | 34% |
| Tango holidays crete greece with leandro furlan | 1.3 MB | 1.1 MB | 18% |
| Tango holidays crete greece with matias guinazu | 1.2 MB | 1.0 MB | 16% |
| Performances | 1.1 MB | 869 KB | 23% |
| Tango argentino | 874 KB | 706 KB | 19% |
| Tango feast christmas paignton england | 720 KB | 720 KB | 0% |
| Tango workshops dinslaken deutschland | 720 KB | 720 KB | 0% |
| Tango norfolk norwich england 3 | 719 KB | 719 KB | 0% |
| Tango norfolk norwich england 7 | 719 KB | 719 KB | 0% |
| Cookie policy | 698 KB | 698 KB | 0% |
| Calendar | 621 KB | 590 KB | 5% |
| Contact | 592 KB | 592 KB | 0% |
| Every page | 26.1 MB | 17.6 MB | 32.6% |
The images alone came down 67.2%. The whole-page figure is lower because every page also carries text, layout and code, which this work does not touch — and the whole page is what a visitor waits for, so that is the number quoted at the top. Four of these pages offer several image sizes and the browser picks one by screen width, so those rows read the first candidate rather than a guaranteed download.
What was not done
Nothing was rebuilt, and nothing was thrown away
Originals are untouched
The AVIF and WebP files sit alongside the originals, never over them. Deactivate the plugin and the site serves exactly the bytes it did before, because those bytes never moved.
No browser is left behind
Serving uses <picture>.
Anything that cannot read AVIF takes the WebP, and anything that cannot read either takes
the original. Nobody sees a broken image.
The design did not change
Same theme, same layout, same photographs at the same dimensions. Put this against a screenshot from the week before and it should be indistinguishable.
Want the same measurement on your site?
The analyser fetches your real images and re-encodes them, so the figure you get is yours, not an average.
Test your site