toolkit Drop a zip whose filenames came out as nonsense — boxes, stray accented letters or symbols where Japanese, Chinese, Korean, Cyrillic or accented names should be — and it is rewritten here, in this browser tab, with every name readable again. Nothing is uploaded: the archive is read and rebuilt entirely on your own device, and the files inside are copied through byte for byte.
Why zip filenames come out garbled
The zip format has no field saying how its filenames are encoded. Windows Explorer stores them in whatever legacy code page the machine runs — Shift_JIS on Japanese Windows, GBK on Chinese, EUC-KR on Korean, CP866 on Russian and Ukrainian, CP850 in much of Western Europe — and sets no flag. A Mac, a phone or a Windows PC set to another language has to guess, and guesses wrong. The reverse happens too: macOS writes UTF-8 names without the flag, and Japanese Windows then reads them as Shift_JIS. Either way the bytes are intact; only the interpretation is wrong, which is what makes the names recoverable.
How the repair works
Every name in the archive is read. Anything already valid UTF-8 is kept exactly as it is. For the rest, the code pages Windows actually writes are each tried in turn and scored on whether they produce plausible words rather than control characters, box-drawing symbols or half-width katakana. One verdict is made for the whole archive, so a short, ambiguous name follows the certainty its neighbours provide. The names are then written back as UTF-8 with the flag set, which Windows 10 and 11, macOS, Linux, iOS, Android and 7-Zip all read correctly.
When auto-detect gets it wrong
Traditional and Simplified Chinese are byte-compatible enough that a Big5 archive can be read as GBK, and a very short name in any script may fit more than one code page. The dropdown lets you name the code page yourself — pick the one matching the Windows the zip came from. A name that auto-detect could not place in any code page is reported in the result rather than silently guessed at.
Junk files, optionally
A zip that has crossed between a Mac and a Windows PC usually carries the __MACOSX folder and .DS_Store files as well as the garbled names, so the same filters the zip cleanup tool offers are here too, with those two on by default. Untick them if you want the archive otherwise untouched; the filename repair runs either way.
What does not change
File contents are copied byte for byte — no decompression, no recompression. Folder structure is preserved, timestamps are preserved, and a password-protected entry is copied through untouched, since it cannot be inspected without the password. The only thing rewritten is the names.
Frequently asked questions
Is the zip uploaded anywhere?
No. The archive is read, its names decoded and the zip rewritten in your browser via WebAssembly; neither the file nor a list of what is in it leaves your device.
Will this fix a zip made on a Japanese Windows PC?
Yes — that is the most common case. Explorer on Japanese Windows stores names in Shift_JIS (code page 932), which macOS and non-Japanese Windows display as garbage. The repair detects it and rewrites the names as UTF-8.
My zip was made on a Mac and shows garbage on Windows. Does this help?
Yes. macOS writes UTF-8 names but leaves the UTF-8 flag unset, so Windows reads them in its own code page. Running the archive through here sets the flag, after which Windows displays the names correctly.
Can it recover names that another tool already mangled?
Only if the original bytes are still in the archive. If a zip was extracted with wrong names and then re-zipped, the garbage is now stored as genuine UTF-8 and is indistinguishable from an intended name — nothing can undo that.
Does it change the files inside?
No — every entry is copied through unchanged, byte for byte. Only the names are rewritten, plus whatever junk files you leave ticked for removal.
What if I don't know which code page the zip uses?
Leave the dropdown on auto-detect. If the result still looks wrong, pick the entry for the language of the computer the zip was made on; the result reports how many names were rewritten and with which code page.