Compare two versions of any text by line, word or character, side by side or as a unified diff you can save as a patch file. Nothing is uploaded.
Waiting for two texts
Compared line by line — a line that differs at all counts as removed and added.
Paste two texts to compare
Configuration files, API responses, two drafts of the same paragraph — anything that is text.
Text Diff runs entirely on your device — nothing is uploaded.
You’re on the Free plan. Compare plans
Two versions of the same thing, and the question is what changed. A diff answers it properly: the matching lines are found first, and only what is left over is reported as added or removed. That is quite different from comparing the two texts position by position, which reports everything after an inserted line as changed even though none of it is.
There are three granularities and the page always says which one is in use, because the answer to what changed is different at each level. Line is the default and the one that scales: it is what diff does, what a patch is made of, and the only level at which a unified diff means anything. A line that differs by a single character is reported there as one line removed and one added, which is right for configuration files, API responses and code.
Word and character answer the question line mode cannot. A rewrapped paragraph or a single corrected word comes back from a line diff as whole lines replaced, which is true and useless when the line is a sentence. Word mode compares token by token and shows the change inside the line; character mode goes further and compares by grapheme cluster, so an emoji built from several code points counts as one character rather than several. Both carry a lower size ceiling than line mode, because tokenising a document into words or graphemes produces far more items to compare than splitting it into lines, and the tool says so rather than slowing to a halt.
There are two views over the same result, and they come from one calculation so they cannot disagree. Side by side keeps line numbers from both texts, which is what you want when reading. Unified is the familiar patch format with context lines and @@ hunk headers, which is what you want when pasting into a review or saving a .diff file.
Three comparison options change what counts as a difference rather than how the output looks: ignore case, ignore whitespace, and ignore blank lines. They are useful for exactly the situations where a file has been through a formatter or an editor with different settings, and the status line always says when the texts are identical only under the options you have chosen.
Large inputs are handled by stripping the common beginning and end before the real work starts, which is what makes two near-identical thousand-line files fast rather than slow. Beyond a documented size the tool declines with a clear message instead of locking the tab, because the honest failure is better than a frozen page. Nothing is uploaded: both texts stay in the browser, which is what makes it safe to diff a config file that still has credentials in it.
All three, chosen with the Compare by control. Line is the default and matches what git does. Word compares token by token and shows a change inside a line. Character compares by grapheme cluster, so a composite emoji counts as one character.
Because a unified diff is a line format — its hunk headers count lines in each file. In word and character mode the copy action gives you the text with changes marked instead, using the wdiff conventions of square brackets for removals and braces for additions.
The standard patch format: a few lines of unchanged context around each change, hunk headers marking the line numbers in both files, and every changed line prefixed with + or -. It is what git diff prints and what patch applies.
No. They change only what counts as a difference while comparing; the lines shown are always exactly what you pasted. When two texts match only because of an option, the status line says so explicitly.
Open each file and paste its contents into the two panels. There is no upload step because there is no server — the comparison happens entirely in the page, which is the point.
Comfortably into the thousands of lines, helped by skipping the identical beginning and end before doing the real work. Past a documented limit it declines with a clear message rather than freezing the tab, which is the more useful failure.
Because a line diff has no concept of a move — it only knows which lines are present in each version. A block that has shifted appears as removed from one place and added in another. Dedicated move detection is a different algorithm with its own false positives.
No. Both texts stay in your browser, nothing is uploaded or logged, and no part of your content reaches analytics. That is what makes it safe to diff a configuration file that still contains secrets.