THE LAB // MEASURED, NOT PROMISED

WebP
vs AVIF.

Both are still images borrowed from video codecs: lossy WebP is a VP8 keyframe (2010), AVIF is an AV1 frame (2019) inside a HEIF container. WebP's lossless mode is a separate codec, and Easydoor doesn't use it. Every image on this page was encoded by the same libvips that Easydoor runs in production.

The only rule that matters here

Comparing quality 80 in WebP with quality 80 in AVIF measures nothing. They are scales of different codecs: AVIF's number is translated into a quantizer from 0 to 63, inverted.

So nobody here compares nominal quality. Either the file size is fixed and you look at what is left of the image, or the measured quality is fixed and you look at how many bytes it cost.

Lab 1

Same file size

Each image was encoded to fit the same byte budget in all three formats, by a binary search on the quality setting: each format gets the highest quality whose file still fits, so files land at or just under the budget, never above it (unless the table marks it over) — the table shows each one's real size. The budget is given in bits per pixel, the honest unit: it does not depend on resolution. The Easydoor option is different — it shows the files at the settings we actually serve, which are not equal in size.

Lab 2

Same measured quality

The other valid way: fix the quality and compare the weight. Each file was encoded until it reached an MS-SSIM target against the original, and what is compared is how many bytes each codec needed. The first option shows Easydoor's own settings, and the quality they reach.

The metric is biased too. MS-SSIM punishes blocking more than blur, which systematically favors AVIF. Lab 4 shows the difference with your own eyes.

Lab 3

The scales don't talk

Visual proof that the quality number is not comparable. Each point is a quality value; its position shows how many bytes and how much quality it actually produced. The ringed points are the settings Easydoor uses.

Lab 4

Anatomy of the error

The tables say how much each codec errs. This shows where. The maps are the absolute error against the original: black is right, blue is a slight error, yellow and white are gross errors. The amplification is set per image, and the three formats of one image share it — compare them with each other, not with another image.

Lab 5

Where the header costs more than the image

AVIF lives in a HEIF container, with metadata boxes of almost fixed size. In a large image that is irrelevant. In an icon it can be most of the file. WordPress also makes a 150 px thumbnail of every upload — the table shows where that lands.

Closing

The verdict — and why nobody has to choose

So Easydoor serves both

Every WordPress size goes out in WebP on upload and in AVIF right after. The browser takes the first format it can read:

<picture>
  <source srcset="photo.avif" type="image/avif">
  <source srcset="photo.webp" type="image/webp">
  <img src="photo.jpg" alt="…" width="1200" height="800">
</picture>

Nobody on your side configures, converts or picks anything.

What this lab does not prove