Text difference engine
Compare two textsdown to the character.
Your text is never saved and never uploaded. Both sides are compared inside this browser tab, on your own machine. Nothing is sent to a server, nothing is stored, and closing the tab erases it completely — safe for contracts, patient notes and anything else confidential.
No account · No upload · No tracePaste an original and a revision. Every insertion, deletion and edited word is measured and marked, with a similarity score, a per-line breakdown and a unified diff you can copy straight into a commit or a review.
Similarity
—
awaiting input
Added
—
lines inserted
Removed
—
lines deleted
Modified
—
lines edited in place
Unchanged — Modified — Added — Removed — Words — Characters —
Myers O(ND) · nothing is uploaded
⌘/Ctrl⏎ compare AltS swap AltK clear Alt↓/↑ next change Alt1–3 view
A real diff, not a line-by-line check
Naive comparison tools walk both texts together and report everything as different from the first change onwards — insert one line at the top and the whole file lights up. This uses Myers' difference algorithm, the approach behind git diff, which finds the shortest edit script between the two versions, so an inserted block is reported as an insertion and everything after it still matches. Identical prefixes and suffixes are trimmed before the search even starts, which is why large files with small edits compare almost instantly.
Once the line-level changes are known, each pair of changed lines is scored for similarity using bigram overlap. Lines that are clearly revisions of each other get a second, word-level pass so you see precisely which words changed; lines with little in common are reported honestly as a separate deletion and addition rather than being force-matched into a confusing highlight. The ignore options change only what counts as identical — the text shown is always exactly what you pasted.
Text Compare FAQ
Is my text uploaded anywhere?
No. Everything happens in JavaScript on your device — you can confirm it by opening your browser's network tab and running a comparison, where you'll see no request. That's why this tool is built client-side by design rather than convenience: diff tools are exactly where people paste contracts, patient notes and production config.
What does the similarity percentage mean?
It's the proportion of characters the two texts share, measured across matched content. 100% means nothing differs; a low score means the two are largely unrelated. It's a useful at-a-glance figure, not a formal edit-distance metric.
What's the difference between word, character and line granularity?
Word is the default and reads best for prose and code. Character catches changes inside a word — a changed digit in an ID, or a typo — at the cost of noisier highlighting. Line skips the inner pass entirely and reports whole lines as changed, which is fastest on very large files.
Can I use the exported diff with git?
Yes. It's standard unified diff format with three lines of context and proper @ hunk headers, so git apply and patch accept it.
Why does it say "compared at block level for speed"?
The search is bounded so a pathological comparison can't hang your browser. When two texts have almost nothing in common, finding a minimal edit script costs more than it's worth, so that region is reported as a straight replacement instead — which is the honest description of it anyway.
Does it handle large files?
Yes — identical prefixes and suffixes are trimmed first, so a 10,000-line file with a handful of edits compares in well under a tenth of a second. Very large inputs will warn you that rendering, not comparing, is the slow part.
What else can you check?
These tools all work on the same connection and address data — pick whichever question you actually have.