这一页只做一件事：把两段文本放在一起，逐行标出加入、删除和改动的部分；必要时再对改动过的行做逐词标注。文件由选择器或拖放读入当前标签页的内存，比较过程全部在本机完成——页面不发起任何网络请求，断网也能用，直接从磁盘打开同样可用。关闭标签页，内容随之消失。
适合用来核对两份配置导出到底差在哪几行、在应用别人的补丁前先看清改动、比较两版文案或译文，以及确认 CSV / JSON 导出只动了预期的字段。行级差异给出增删的骨架，词级差异指出同一行里真正变化的词；并排与统一两种视图、连续相同行的折叠，是为长文件准备的。
比较只在文本上做：二进制、加密或压缩文件会被识别并明确拒绝，而不是给出半截结果。需要格式转换（Markdown → Word、Excel → CSV、图片格式互转等）时，请用同一作者的 Transfer Any File 工作台，或从按格式分类的转换页进入。两者都在本地完成，不上传文件。
不会。文件只在当前标签页的内存里读取和比较，页面不发起任何网络请求，断网也能用；关闭标签页后内容消失。直接从磁盘打开这个 HTML 文件时，行为相同。
每一边上限 32 MiB、40 万行，两边要同时放进内存。超过上限会明确说明原因并停止，不会截断成半份结果。
行级先确定哪些行被加入、删除或改动；对改动过的行，词级再标出真正变化的词。超长行会退回行级，并在结果里注明。
能。复制或下载得到的是标准 unified diff，含 3 行上下文与 @@ 区间，可直接交给 git apply；导出的补丁统一使用 LF 换行。
This page does one thing: put two pieces of text side by side and mark every added, removed or changed line, then mark the words that truly differ inside a changed line. Files are read into this tab’s memory by the picker or by drag and drop, and the comparison runs entirely on your machine. The page makes no network request, works with the network off, and behaves the same when the HTML file is opened straight from disk. Close the tab and the content is gone.
It is for checking which lines two exports of a config actually differ in, reading a change before applying someone’s patch, comparing two revisions of copy or a translation, and confirming that a CSV / JSON export touched only the fields you expected. Line-level output gives the skeleton of additions and removals; word-level output points at the part of a line that really changed. Split and unified views, plus folding of long unchanged runs, are there for big files.
The comparison is textual only: binary, encrypted or archived files are recognised and refused outright rather than half-read. For format conversion — Markdown to Word, Excel to CSV, one image format to another — use the Transfer Any File workbench from the same author, or enter through the format-by-format conversion pages. Both run locally and upload nothing.
No. They are read into this tab’s memory and compared there; the page issues no network request, works offline, and discards the content when the tab closes. Opening the HTML file straight from disk behaves the same way.
Up to 32 MiB and 400,000 lines per side, with both sides held in memory at once. Exceeding a limit stops with a stated reason rather than returning half a result.
Line level decides which lines were added, removed or changed. For changed lines, word level marks the words that actually differ. Very long lines fall back to line level and say so in the result.
Yes. Copy or download produces a standard unified diff with three lines of context and @@ hunks, ready for git apply; exported patches always use LF line endings.