Back to Top Start a Trial

Small-business homepage image weight

Origin images from 19,522 small-business homepages re-encoded in PicPerf's analyzer: median payload 0.88 MB, median remaining 40.5%, median 16 images. Untreated census, not field RUM.

By Alex MacArthur, builder of PicPerf · Published

Headline (2026-09-07): across 19,522 unique small-business homepages, the median page carried 0.88 MB of origin images. After this encode recipe, the median remaining size was 40.5% of the original bytes. The median page loaded 16 images. These are origin images re-encoded in PicPerf's analyzer. Not field RUM, and not a product before-and-after.

Median image payload
0.88 MB
19,522 unique homepages
Median remaining after encode
40.5%
≈59% smaller than original
Median image count
16
8 p25 · 26 p75
Image share by original format
Share of all analyzed images, by the content type served on the origin. Longer bars are more common.
Remaining size buckets
URLs by remaining bytes after this encode recipe, as a share of the original. Shorter remaining is better.

Image weight at a glance

One row per unique URL, latest analysis kept. PicPerf page image analyses (crawl + encode). Latest analysis per unique URL.

Metric n p25 Median p75
Image payload (MB) 19,522 0.33 0.88 2.2
Remaining bytes (MB) 19,522 0.14 0.31 0.61
Remaining (% of original) 19,522 22.5% 40.5% 60.3%
Image count 19,522 8 16 26

Remaining-size buckets

Share of URLs by remaining bytes after this encode recipe, plus the share of pages carrying 1 MB or more of origin images.

Remaining URLs Share
< 50% 12,154 62.3%
50–89% 6,425 32.9%
≥ 90% 943 4.8%
≥ 1 MB original payload 9,032 46.3%

Outreach vs public analyzer

Some URLs come from PicPerf's outreach pipeline; the rest come from people who ran the public image analyzer. Promo jobs cap at 15 images per analysis.

Source n Median remaining Median payload Median images
Outreach (promo) 4,969 38.7% 1.01 MB 16
Public analyzer 14,553 41% 0.84 MB 16

By original format

Every analyzed image by the content type served on the origin. Median original size is per image; median remaining is per image after this encode recipe.

Format Images Share Median size Median remaining
PNG 115,010 29.8% 14.6 KB 41%
JPEG 96,199 24.9% 56.8 KB 47.1%
WEBP 82,110 21.3% 27.3 KB 79.3%
SVG 54,127 14% 2.8 KB 98.3%
AVIF 22,848 5.9% 5.1 KB 59.2%
OTHER 10,945 2.8% 25.9 KB 60.3%
GIF 4,343 1.1% 8.6 KB 62.2%

Methodology

  1. Each analysis crawls a homepage's images and re-encodes them with PicPerf's image pipeline (modern output format when smaller, optional downscale to the analysis width).
  2. Duplicate URLs keep the latest analysis only.
  3. Outreach (promo) jobs cap at 15 images per analysis; the public analyzer does not.
  4. The encode recipe includes an optional resize (promo jobs often cap width at 1100px), so remaining bytes reflect resizing plus re-encoding, not the encoder alone.
  5. The encoder was not constant over 2024–2026: the output format at analysis time varied (WebP, AVIF, and JXL historically).
  6. This dataset includes no per-site URL list.

Machine-readable tables: JSON and CSV. Page-speed lab baselines for a different homepage sample are a separate dataset: small-business homepage lab performance.

What this does not claim

  • Not the whole web. This is a convenience sample of sites PicPerf analyzed. It is complementary to HTTP Archive, not a replacement.
  • Not field data. These are re-encoded origin bytes, not CrUX or RUM.
  • Not the HTTP Archive byte index. Different sample, different recipe.
  • Not live PicPerf CDN traffic. Nothing here measures production image_requests.
  • Not Turbulence HTML rewrites. This dataset belongs to PicPerf Images, not the full-site product. Paired before-and-after product results live at Turbulence performance research.
  • Remaining ratio is not “Lighthouse wastedBytes.” Lighthouse image audits estimate per-page waste under their own assumptions; this is a straight re-encode census.

Frequently Asked Questions

  • In this sample of 19,522 unique homepages (snapshot 2026-09-07), the median page carried 0.88 MB of origin images. The 25th percentile was 0.33 MB and the 75th percentile was 2.2 MB. 9,032 URLs (46.3%) carried 1 MB or more. These are untreated origin images, not field RUM.

  • After this encode recipe, the median remaining size was 40.5% of the original bytes (p25 22.5%, p75 60.3%). 12,154 URLs kept less than half their original bytes. That is not a pure encoder saving: the recipe includes optional downscaling (promo jobs often cap width at 1100px), and the output format varied over time (WebP, AVIF, and JXL historically).

  • The median homepage in this sample loaded 16 images (p25 8, p75 26). Promo-pipeline jobs cap at 15 images per analysis, so counts above 15 come from the public analyzer.

  • Mostly not at the origin. Of all analyzed images, PNG was 29.8% and JPEG 24.9%, while WebP was 21.3% and AVIF only 5.9%. SVG was 14.0% (median 2.8 KB each). The formats with the most headroom were the legacy ones: PNG images kept a median 41.0% of bytes after re-encoding, JPEG 47.1%.

  • Neither. Every figure comes from origin images re-encoded in PicPerf's analyzer, frozen as a static snapshot. It is an untreated image-payload census, not CrUX field data, not the HTTP Archive byte index, not live PicPerf CDN traffic, and not Turbulence HTML rewrites.

Related Guides

Prefix image URLs or upload originals. AVIF/WebP when smaller, resizing, and a 1-year CDN cache.

Start a 14-day Image Trial