Skip to content
EstimatorBy 7 min read

How much storage do daily screenshots take?

The same 100 captures of the same pages, at 1280 px wide, weighed 10.3 MB one way and 168 MB another: about 16 times apart. Below are the six ways, and an estimator that turns your own numbers into gigabytes.

  • First screen, WebP10.3 MB
  • First screen, JPEG15.5 MB
  • First screen, PNG23.1 MB
  • Whole page, WebP61.0 MB
  • Whole page, JPEG90.4 MB
  • Whole page, PNG168 MB

Mean of seven real pages, measured in Chromium on 12 November 2026. Each bar is 100 pictures of one kind.

The estimator

Widths, in px

The average is the mean of the seven pages measured. Pick one to see a single page.

Starts at the Cloudflare R2 Standard price, read on 12 November 2026. Change it to your provider.

Your own number: how often a page looks exactly like its last capture. It only changes the lower line of each result.

50 pages × 1 a day × 2 widths = 100 pictures a day, about 80.3 KB each

New data a month

241 MB

with dedupe: 72.3 MB

New data a year

2.93 GB

with dedupe: 879 MB

Held at once, once 90 days have passed

723 MB costs $0.01 a month

with dedupe: 217 MB, < $0.01 a month, 506 MB less

3,000 screenshots over 30 days. The smallest plan that holds them is Starter (3,000 a month). Dedupe does not lower this number: an unchanged page is still captured.

Held at once, by format (same inputs)

  • PNG1.56 GB
  • JPEG, quality 801.05 GB
  • WebP, quality 80723 MB

Every size in the estimator is a measured file size from the tables below. A GB is 10^9 bytes and a month is 30 days. The held-at-once figure is what is on disk once the keep period has passed and the oldest pictures are being deleted. Plans and prices.

Seven pages, weighed

One capture of each page at 1280 px wide, kept at quality 80 for JPEG and WebP. The first screen is 1024 px tall, which is what a request without full_page returns. The sizes run from 40.7 KB to 3.96 MB at this one width.

First screen

PagePNGJPEGWebP
example.comA nearly empty page86.3 KB62.2 KB40.7 KB
Hacker News front pageA plain text list232 KB142 KB100 KB
MDN: CSS referenceDocumentation, mostly text194 KB205 KB138 KB
GitHub repository pageA web app with a long README222 KB164 KB99.7 KB
Wikipedia: Web serverA very long article392 KB193 KB133 KB
BBC News front pageNews with many photos253 KB148 KB96.7 KB
Wikipedia: Featured picturesA photo gallery241 KB169 KB112 KB

Whole page

PageHeightPNGJPEGWebP
example.comA nearly empty page1,024 px86.3 KB62.2 KB40.7 KB
Hacker News front pageA plain text list1,183 px248 KB157 KB108 KB
MDN: CSS referenceDocumentation, mostly text4,274 px532 KB549 KB369 KB
GitHub repository pageA web app with a long README8,527 px1.19 MB936 KB568 KB
Wikipedia: Web serverA very long article16,000 px3.96 MB2.78 MB1.95 MB
BBC News front pageNews with many photos6,630 px2.91 MB1.04 MB679 KB
Wikipedia: Featured picturesA photo gallery6,905 px2.81 MB808 KB555 KB

Wikipedia: Web server is 29,470 px tall at this width, so its whole-page row is a capture cut at 16,000 px. The cut is explained under Format and quality.

Width moves the size less than you think

A phone picture has fewer pixels, but a phone page is longer, so a whole-page capture at 390 px is not small. The mean of the seven pages at each width:

WidthScreen PNGScreen JPEGScreen WebPPage PNGPage JPEGPage WebP
390 px116 KB79.4 KB57.6 KB1.01 MB573 KB412 KB
768 px174 KB116 KB80.5 KB1.53 MB809 KB571 KB
1280 px231 KB155 KB103 KB1.68 MB904 KB610 KB
1920 px250 KB170 KB109 KB1.80 MB1.00 MB656 KB

Going from 1280 to 1920 px added 6% to a WebP first screen. Going from a first screen to a whole page multiplied it by 5.9. The choice that saves the most is whether you need the whole page. See the format comparison for what each format is for.

Format and quality

WebP

Smaller than JPEG in 56 of 56 cases

At quality 80 it was 20% to 96% of the PNG. The best case was the photo gallery, the worst the documentation page.

JPEG

Not smaller than PNG in 6 of 56 cases

At quality 80 it was 28% to 125% of the PNG. It did not beat PNG on MDN: CSS reference and BBC News front page, pages with large areas of flat colour and text, where PNG compresses well.

PNG

Largest on every photo page

The whole-page PNG of the photo gallery was 2.81 MB, against 555 KB as WebP. Keep PNG when you need a lossless picture or a transparent one.

Quality 60 and 90, against quality 80

Mean of the seven pages at 1280 px wide. The image_quality option takes 1 to 100 and defaults to 80.

Format and captureQuality 60Quality 90
JPEG, first screen72%135%
JPEG, whole page72%135%
WebP, first screen78%135%
WebP, whole page77%136%

A WebP file cannot be longer than 16,383 px

Wikipedia: Web server is 29,470 px tall at 1280 px wide. At 20,000 px the WebP encoder refused the picture, so the whole-page captures of this page in the run are cut at 16,000 px. For a page this long, use JPEG, or cap it with full_page_max_height.

Do a page's repeats match?

Dedupe works when a capture of an unchanged page is the same file as the last one: a hash of the file then tells you to drop it. Whether it is the same file is something to test, so each page was captured again.

Page3 captures, 4 seconds apart5 captures, 3 minutes apart
example.comall the same fileall the same file
Hacker News front pageall the same file4 different files
MDN: CSS referenceall the same fileall the same file
GitHub repository pageall the same file2 different files
Wikipedia: Web servernot run5 different files
BBC News front pageall the same file5 different files
Wikipedia: Featured picturesall the same fileall the same file

3 of 7 pages gave the same file every time over 12 minutes. 4 gave a new one at least once: Hacker News front page, GitHub repository page, Wikipedia: Web server, BBC News front page. A change of one pixel is a different file, so a live page, a clock, a rotating banner or a test variant gives a new file on every capture. The share of captures that repeat over days and weeks was not measured, and no table here can give it: it depends on your pages. That is why the estimator asks for it. Set it to 0 to see the worst case.

keep.sh
#!/usr/bin/env bash
# keep.sh NEW_FILE DIR: keep NEW_FILE in DIR only when it differs from the last kept picture.
set -euo pipefail
new=$1
dir=$2
mkdir -p "$dir"
hash=$(sha256sum "$new" | cut -d' ' -f1)
if [ -f "$dir/latest.sha256" ] && [ "$hash" = "$(cat "$dir/latest.sha256")" ]; then
  rm "$new"
  echo "same as last, dropped"
else
  mv "$new" "$dir/$(date +%F-%H%M%S).png"
  echo "$hash" > "$dir/latest.sha256"
  echo "kept"
fi

The script was run four times on the lab host with small files: the same bytes twice (the second was dropped), then different bytes (kept), then the first bytes again (kept, because it differs from the last kept picture). It passed bash -n. Run it once for each page and width, with its own folder.

Dedupe saves storage, not screenshots

Dropping a repeat happens after the render, so the capture was already taken and counted against your plan. To use fewer screenshots, capture less often; see change monitoring with screenshots.

Three setups, worked out

The same arithmetic as the estimator, with storage at $0.015 per GB-month. Open the estimator and enter any of them to check.

Landing page watch

20 pages, once a day, desktop and phone, WebP first screens, kept 30 days. 80% of captures identical to the last.

Screenshots a month
1,200
Smallest plan
Starter
New data a month
96.4 MB
Held at once
96.4 MB
Held, with dedupe
19.3 MB
Storage a month, with dedupe
< $0.01

Agency, client sites

200 pages, once a day, desktop and phone, JPEG whole pages, kept 90 days. 60% of captures identical to the last.

Screenshots a month
12,000
Smallest plan
Growth
New data a month
8.86 GB
Held at once
26.6 GB
Held, with dedupe
10.6 GB
Storage a month, with dedupe
$0.16

Proof archive

100 pages, four times a day, desktop, PNG whole pages, kept 2 years. 50% of captures identical to the last.

Screenshots a month
12,000
Smallest plan
Growth
New data a month
20.1 GB
Held at once
490 GB
Held, with dedupe
245 GB
Storage a month, with dedupe
$3.67

The storage price is the Cloudflare R2 Standard rate, with the first 10 GB-month free each month [1]; your provider may differ. One successful render counts as one screenshot of your plan whatever its size, so a bigger file never uses more of it: Free has 100 screenshots a month, Starter 3,000, Growth 15,000 and Scale 60,000. See the plans. To run a list like this on a schedule, see the bulk screenshots tool or the screenshot API, and the guide to screenshotting a list of URLs.

How it was measured

Seven public pages were opened in Chromium at four widths (390, 768, 1280 and 1920 px) with a window 1024 px tall, scale 1. Each page was scrolled down and back to load lazy pictures, then captured twice: the first screen (1024 px tall) and the whole page (cut at 20,000 px, or 16,000 px for the long article). Each capture was written the way the renderer writes it: PNG straight from Chromium, JPEG straight from Chromium at the quality given, and WebP re-encoded from the PNG at the quality given. The size of every file was recorded in bytes. Quality 80, the default of the image_quality option, was used everywhere; 60 and 90 were added at 1280 px. That is 56 pairs of pictures, in three formats each. Tools: Chromium 153.0.8010.12, Playwright 1.63.0, sharp 0.35.5 (libwebp 1.6.0). Run on 12 November 2026.

The first run of the Hacker News page at 1920 px produced a first screen of 15 KB on a page only 1,024 px tall, against about 230 KB at the other widths. I did not look at it; it looks like an error or block page after many requests. The page was captured again at that width and the second run is the one used. The three-captures and five-captures checks used the first screen at 1280 px as PNG.

Seven pages are a small sample, and a page changes: the same address weighs something different next month. Scale 2 (retina) captures, pages behind a login, transparent PNGs and video were not measured. The figures are the size of the file on disk, before any compression your storage provider applies, and before the extra cost of backups or copies.

Questions people ask

How much storage does one website screenshot take?

It depends on the page, the width and the format. At 1280 px wide the mean of seven real pages was 103 KB as a WebP first screen and 1.68 MB as a PNG whole page, measured on 12 November 2026. Single files in the run ranged from 37.3 KB to 4.32 MB.

Which screenshot format uses the least storage?

In this run WebP at quality 80 was the smallest in 56 of 56 cases against JPEG at quality 80, and 20% to 96% of the PNG. WebP files cannot be longer than 16,383 px, so a very long page needs JPEG or a lower full_page_max_height.

How do I estimate monthly screenshot storage?

Multiply pages, captures per page a day and widths to get pictures a day, multiply by the average size of one picture, then by 30 for a month or by the days you keep each file for the amount held at once. The estimator on this page does it from the measured sizes.

Does deduplicating screenshots save space?

Only for pages that do not change. In the lab, captures of the same unchanged page came out as identical files, so a hash of the file finds repeats. 4 of 7 pages gave a different file at least once in five captures over 12 minutes. Deduplication saves storage, not screenshots: each capture still counts on your CurlShot plan.

Does a bigger screenshot use more of my plan?

No. On CurlShot, one successful render is one screenshot whatever its size, and failed renders and cache hits are not counted. Only the storage you keep grows with file size.

Screenshot any page in one step

Paste an address into CurlShot and download the screenshot. A free account gives you 100 screenshots every month. No card needed.