Skip to content
Myth listBy 6 min read

Six screenshot myths, tested in Chromium

Six things people say about website screenshots. Each one was drawn on a test page, and the picture was read back. Here is the scoreboard; the proof is below it.

  1. The same page gives the same picture every time.Partly true
  2. The picture is as wide as the viewport.Breaks
  3. Chromium cannot draw a picture taller than 16,384 px.Partly true
  4. A retina picture is twice the file size.Breaks
  5. omit_background makes any page transparent.Partly true
  6. A full-page picture is the same as a very tall viewport.Breaks
Myth listRun on 9 November 2026Chromium 153

Partly true

“The same page gives the same picture every time.”

True for a still page: 1 file in 30 captures. False as soon as something moves: up to 30 different files in 30.

Each of five test pages was drawn 30 times, each time in a new browser context. The files were compared by SHA-256, so one changed pixel counts as a different file.

What to do. Do not compare captures byte for byte. Turn animation off with reduced_motion, wait for the element you care about with wait_for_selector, and compare with a tolerance. The article on visual regression testing shows how.

Different files in 30 pictures

  • A still page: text and a gradient1
  • A focused text field with a caret1
  • A box that spins forever3
  • A block that fades in over half a second4
  • A line that prints the time of the load30

The caret of a focused field did not change the file in this run. The spinner and the fade changed it with the moment of the capture.

Breaks

“The picture is as wide as the viewport.”

A full-page picture of a page with one 2,000 px block came out 2,000 px wide in a 1280 px viewport, and 4,000 px wide at scale 2. Hiding the overflow did not change it.

The viewport was 1280 px wide and every picture was taken with full_page on.[1] Playwright describes it as a screenshot of the full scrollable page.

PagePicture widthPicture heightWider than the viewport?
A block 2,000 px wide2,000 px1,043 pxYes
The same, overflow-x: hidden on body2,000 px1,043 pxYes
The same, overflow-x: hidden on html2,000 px1,043 pxYes
The 2,000 px block at device_scale_factor 24,000 px2,086 pxBoth sides doubled
A page whose blocks are 100% wide1,280 px2,000 pxNo

What to do. Find what is wider than the screen (document.documentElement.scrollWidth tells you), fix it, or ask for a fixed area with clip_width.

Partly true

“Chromium cannot draw a picture taller than 16,384 px.”

False for PNG and JPEG: 40,000 px tall came out whole. True for WebP: it stops at 16,383 px, and above that the file was empty.

One gradient page, 400 px wide, drawn full page at seven heights in three formats. A cell shows the height of the picture that came back. “Empty file” means the browser returned no data at all, not an error picture.

Page heightPNGJPEGWebP
8,000 px8,000 px8,000 px8,000 px
16,383 px16,383 px16,383 px16,383 px
16,384 px16,384 px16,384 pxEmpty file
16,385 px16,385 px16,385 pxEmpty file
20,000 px20,000 px20,000 pxEmpty file
30,000 px30,000 px30,000 pxEmpty file
40,000 px40,000 px40,000 pxEmpty file

At device_scale_factor 2, a 16,384 px page came back as a 32,768 px PNG. The WebP line matches the format, which allows at most 16,383 x 16,383 px.[2] The quiz article asks what the API returns for a tall WebP: URL to image quiz.

What to do. Use png or jpeg for a tall page. The API cuts a full-page capture off at 20,000 px unless you set full_page_max_height, which goes up to 30,000.

Breaks

“A retina picture is twice the file size.”

It has four times the pixels at scale 2 and nine times at scale 3. The files grew 1.9 to 3.9 times at scale 2, depending on the page and the format.

Four 1280 x 800 test pages, each drawn at scale 1, 2 and 3 as PNG and as JPEG at quality 80. The bars compare the PNG at each scale with the longest bar on this table.

Text page

  • Scale 1 PNG121 KB
  • Scale 2 PNG320 KB2.6x
  • Scale 3 PNG524 KB4.3x

JPEG: 139 KB, 362 KB (2.6x), 574 KB (4.1x).

Flat colour

  • Scale 1 PNG4.7 KB
  • Scale 2 PNG15.6 KB3.3x
  • Scale 3 PNG32.7 KB6.9x

JPEG: 6.8 KB, 24.8 KB (3.7x), 54.8 KB (8.1x).

Gradient

  • Scale 1 PNG185 KB
  • Scale 2 PNG719 KB3.9x
  • Scale 3 PNG1,489 KB8.1x

JPEG: 15.4 KB, 38.6 KB (2.5x), 69.9 KB (4.5x).

900 coloured circles

  • Scale 1 PNG265 KB
  • Scale 2 PNG939 KB3.5x
  • Scale 3 PNG1,608 KB6.1x

JPEG: 114 KB, 222 KB (1.9x), 358 KB (3.1x).

At scale 2 the files grew 1.9x to 3.9x; the pixels grew 4 times, and the byte count follows how well the format squeezes that page. See PNG, JPEG, WebP or PDF for choosing the format.

What to do. Use scale 2 where the picture is shown small and sharp, not by default. Of the four test pages, the one with many circles grew least as JPEG, 1.9 times. One render counts as one screenshot whatever its size, but a bigger file takes longer to send.

Partly true

“omit_background makes any page transparent.”

It removes the browser default white and nothing else. A bare page came out fully transparent; a page that paints its own white came out with 0 percent transparent pixels.

Five tiny pages in an 800 x 400 viewport, drawn once with omit_background and once without. Playwright says the option hides the default white background and does not apply to JPEG.[1] The table gives the share of pixels that were not fully opaque.

PageOption onOption off
A bare page, no background set100% transparent0% transparent
body { background: #fff }0% transparent0% transparent
html { background: #fff }0% transparent0% transparent
A full-height wrapper with a white background0% transparent0% transparent
A white 400 x 200 card on a bare page75.1% transparent0% transparent

The card case is the useful one: the white card stayed, and everything around it became clear, which is what you want for a component on a transparent canvas. The options page lists the formats that can hold transparency.

What to do. Check what paints the background: body, html or a wrapper. Your own page can clear it; a page you do not own stays opaque. Use png or webp.

Breaks

“A full-page picture is the same as a very tall viewport.”

On a page made of three 100vh sections, a 2,400 px viewport made the page 7,200 px tall, and the plain picture showed one third of it.

A test page of three sections, each height: 100vh. Playwright describes a full-page screenshot as one taken as if you had a very tall screen.[1] On a page that sizes itself in vh, a very tall screen is a different page.

What to do. Keep a normal viewport and set full_page. If a hero is 100vh, expect it to be one screen tall in the result.

Picture and page height, in px

  • Viewport 1280 x 800, full page

    Picture2,400
    Page (scrollHeight)2,400
  • Viewport 1280 x 2400, no full page

    Picture2,400
    Page (scrollHeight)7,200
  • Viewport 1280 x 2400, full page

    Picture7,200
    Page (scrollHeight)7,200

With the tall viewport and no full page, the picture is 2,400 px of a page that grew to 7,200.

How it was run

On 9 November 2026. Test pages were drawn by Chromium 153.0.8010.12 through Playwright, on a Linux server, with a new browser context for every picture. Each picture was read back from the file: its size from the header, its pixels with sharp, its identity as a SHA-256 hash.

Every test page is small and made for one question, so the numbers show what Chromium does with that page, not with yours. Real pages mix many of these effects. Counts of 30 pictures per page for the first myth are small, and the spinner and the fade would give other counts on another run; only the shape of the result is the point.

The options named in the fixes (reduced_motion, wait_for_selector, clip_width, full_page_max_height, device_scale_factor, omit_background) come from the documentation of CurlShot, and the cut-off of 20,000 px and its maximum of 30,000 px were read there on 9 November 2026. No call to the API was sent in this run, so none of the pictures above came from the service. Try any of them on the full-page screenshot tool or the high-resolution screenshot tool, which need no account, and see pricing for the free plan.

Why these six

They are claims that look right and cost a retake when they are wrong. For the bigger picture of why a capture can differ from what you see in your own browser, read why website screenshots look wrong.

Questions people ask

Is a full-page screenshot the same as a very tall viewport?

No. In a run on 9 November 2026, a page of three 100vh sections was 2,400 px tall at a viewport of 1280 x 800 and 7,200 px tall at 1280 x 2400, because 100vh grew with the viewport. A normal viewport with full-page capture shows the page as a visitor scrolls it.

Why is my full-page screenshot wider than the viewport?

Something on the page is wider than the screen, such as a table, a code block or an image with a fixed width. A full-page picture covers the whole scrollable area, and in the run here, hiding the overflow on body or html did not shrink it. Find the wide element and fix it, or capture a fixed area.

How tall can a full-page screenshot be?

In Chromium, PNG and JPEG came out whole at 40,000 px in the run on 9 November 2026. WebP is limited to 16,383 px on a side by the format. CurlShot cuts a full-page capture at 20,000 px by default and allows up to 30,000 with full_page_max_height.

Does a retina screenshot double the file size?

No. At device_scale_factor 2 the picture holds four times the pixels, and the files in the run grew from 1.9x to 3.9x, depending on the page and on whether it was PNG or JPEG.

Why is my screenshot background not transparent?

omit_background removes the browser default white. If the page, its body or a wrapper paints its own background, that stays. A bare page came out fully transparent in the run, and a page with a white body came out with no transparent pixels. Use png or webp.

Sources

  1. [1]Playwright, Screenshots: full page screenshots. Checked .
  2. [2]Google for Developers, WebP FAQ: maximum pixel dimensions. Checked .

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.