Full page capture

The browser will only ever give an extension the pixels currently on screen. There is no API that hands over the whole document as an image, so every full page capture — in every tool, including Chrome's own — is a scroll-and-join underneath.

The interesting part is the bookkeeping between those captures.

What the engine does per step

Measures instead of assuming

After each scroll the engine reads the actual position back from the page. Browsers clamp and round scroll offsets, and the final step before the bottom never moves a full viewport, so an assumed distance duplicates a strip of content at the join.

Trims from the measurement

The overlap trimmed off each new capture is computed from how far down the output is already filled against where the page really landed. The join is exact rather than approximately right.

Works in device pixels

Capture happens at the display's real density. On a 2× screen you get a 2× image, and the crop maths runs in device pixels so fractional zoom levels do not accumulate error down the page.

Frees memory as it goes

Bands that are complete are encoded and released rather than held until the end, so peak memory stays roughly constant whether the page is 3,000 or 40,000 pixels tall.

Restores everything

Every change made to the page is recorded and undone, including the original scroll position. A watchdog restores the page even if the capture is interrupted.

Capture characteristics

Pixel density

Native — 1×, 2× or 3×

Overlap between bands

Trimmed to zero from measured positions

Peak memory

Bounded — bands flushed as completed

Page state after capture

Restored, including scroll position

Interruption handling

25-second watchdog restores the page