Comparing loads two versions of a document side by side, page by page, with an overlay that highlights the areas that differ in red. It answers the question that comes up whenever a contract, a specification or a policy comes back from someone else: what actually changed? Reading two long documents in parallel is slow and unreliable — the human eye is very good at seeing what it expects — while a visual diff surfaces a changed figure or an inserted clause immediately. The comparison is visual rather than semantic: it renders both pages and looks for regions that do not match, which catches everything from a reworded sentence to a moved table, but flags reflowed text as changed even when the words are identical. Both files are processed in your browser, so two versions of a confidential agreement are never uploaded.
How the difference is computed
Both pages are rendered to bitmaps at the same scale and compared region by region; areas that do not match are tinted. This is deliberately dumb and therefore complete — it cannot miss a change, because any visible change alters the pixels.
The trade is that it does not understand what changed. A paragraph inserted on page 3 pushes everything after it down, and every subsequent page then reads as almost entirely different, even though the text is unchanged. That is the main thing to be ready for when comparing documents that have grown.
Reading the result
Work page by page and treat the highlights as candidates rather than conclusions. Small isolated marks are usually the real edits: a changed number, a new clause, a different date, a swapped name.
Large uniform highlighting across a page usually means reflow rather than rewriting. When you see it, find the first genuine change above it — that is the insertion or deletion that shifted everything, and the pages after it are often unchanged in substance.
What this is good and bad at
It is good at documents from the same source with the same layout: two exports of a contract, two revisions of a drawing, a policy before and after amendment. It catches everything, including changes to images, tables and figures that a text diff would miss entirely.
It is poor at documents that were regenerated with different fonts or margins, and at comparing a scan to a digital original, where nothing lines up and everything highlights. For those, extracting the text from both and comparing that is the more useful approach.