Full-page screenshots came out wrong or failed on a few kinds of pages. They now render.
Pages that scroll inside an element
Some web apps don’t scroll the window. They scroll a container that fills the screen, for example, a dashboard with a fixed sidebar and a scrolling content area. A full-page screenshot of such a page showed only one screen.
Now the API finds that container and captures all of its content. Modals, drawers and other fixed layers are left as they are.
Full-page screenshots of these pages are therefore taller than the ones you took before.
Pages that replace the browser’s scrolling functions
Some sites define their own scrollTo or scrollBy function, which replaces the browser’s. On these pages:
- lazy-loaded content didn’t load in full-page screenshots;
scroll_into_viewran until the request timed out;- screenshots with
full_page_algorithm=by_sectionsfailed with an internal error.
The API no longer depends on these functions, and all three work.
Very large pages
When a page is both wider than the viewport and too large to capture at its own width, the request failed. It is now captured at the viewport’s width instead.
A page that is still too large for the image format returns resulting_image_too_large, as before. Use full_page_max_height to limit its height.
Repeated rows with a maximum height
With full_page_algorithm=by_sections and full_page_max_height, the bottom of the image could show a few rows of the page twice. It doesn’t anymore.
Pages that load another page during the capture
Some pages replace themselves after they load, for example, by a redirect from a script or a reload. When that happened while the API was capturing the page by sections, the request failed with an internal error, or the image was stitched from two different pages.
Now the API waits for the new page and captures it from the start. If the page keeps changing, you get a network_error.


