ファイルを選ぶ
ドラッグ&ドロップ、またはクリックして選択します。複数ファイルの同時処理が可能で、サイズ制限はありません。ファイルはメモリーに読み込まれ、どこにも送信されません。
数秒でどんなファイルも破損させる、 どこにもアップロードせずに
ランダムなバイトを数個書き換えるのではなく、実際のパーサーが依存する構造そのものを破壊する、フォーマットを理解した破損ツールです。PDF、Office 文書、アーカイブ、動画、音声、画像に対応し、すべてブラウザー内で処理します。
Everything runs locally. Nothing is uploaded, nothing is stored.
どの形式にも専用のプロファイルがあり、そのパーサーが依存する構造に合わせて調整されています。
Office documents and PDFs are the formats people most often need broken, and the hardest to break convincingly.
10 個のツールText-based formats have no header to destroy, so scrambling characters just produces different text.
7 個のツールMedia formats are built to survive damage: players skip broken regions and image decoders resynchronise after a bad chunk.
7 個のツールArchives are the formats most explicitly engineered against corruption.
5 個のツールExecutables and installers are validated before anything runs, which makes their failure modes precise and reproducible.
4 個のツールNot everything has to be serious.
2 個のツール3 ステップ、アカウント不要、待ち時間なし。
ドラッグ&ドロップ、またはクリックして選択します。複数ファイルの同時処理が可能で、サイズ制限はありません。ファイルはメモリーに読み込まれ、どこにも送信されません。
ほとんどの用途では Standard が適切です。エンジンが拡張子とマジックバイトから形式を判別し、その形式のプロファイルを適用します。
破損したコピーはメモリー上で生成され、元のファイル名でダウンロードされます。元のファイルは一切変更されません。
Most online file corrupters overwrite a few dozen random bytes. On a plain text file that is enough. On the formats people actually need to break it usually is not, because those formats are containers built to survive damage.
The engine reads the extension and the magic bytes, picks one of 87 format profiles, and destroys that format's structural landmarks before it touches the payload. Random byte flips usually land inside a compressed stream, which is exactly the damage Word and Acrobat are designed to recover from.
Files are read with the File API, rewritten in memory and handed back through a Blob URL. There is no upload endpoint in the codebase, so there is no server log with your filename in it, no retention policy to trust and nothing to breach.
Light leaves a file that opens with a warning or renders partially, which is what produces glitch art. Standard makes it unopenable. Heavy defeats repair tools. Shred replaces every byte with cryptographic noise while keeping the name and size.
The corrupted copy keeps the original filename, the exact byte length and the correct MIME type. It copies, syncs, attaches to email and passes upload forms like any other file. It only fails at the moment something tries to parse it.
Processing runs in 8 MB slices with a yield to the event loop between them, so a multi-gigabyte video neither freezes the tab nor exhausts memory. Drop in a folder and every file is processed in one pass, at roughly 18 ms per megabyte.
Every run reports the profile used, how many structure markers were destroyed, the percentage of bytes rewritten and a hex view of the header before and after. Optional SHA-256 hashing proves the input and output differ.
Header, index and trailer are gone. Nothing can locate a single entry, so repair has nothing to rebuild from.
Corruption is applied to structure first and payload second. This is the landmark each profile targets, and the error the application actually shows afterwards.
Acrobat: "The file is damaged and could not be repaired."
The reader opens a PDF from the back, so every index at the tail has to go.
| Format | Structure destroyed | What you see |
|---|---|---|
The %PDF- header, every xref table, the trailer and the %%EOF marker | Acrobat: "The file is damaged and could not be repaired" | |
| DOCX, XLSX, PPTX | Local file headers, the central directory and the end-of-central-directory record | Word: "unreadable content", and Open and Repair then fails |
| ZIP, RAR, 7z | The archive index, local headers and RAR recovery records | WinRAR: "Cannot open file as archive" |
| ISO | The full 32 KB system area and the CD001 volume descriptors | Windows: "There was a problem mounting the file" |
| DMG | The koly trailer and the mish block map | macOS: "No mountable file systems" |
| MP4, MOV, MKV | The ftyp brand and the moov atom holding the sample table | VLC: "Your input can't be opened" |
| MP3, WAV, FLAC | ID3 tags, MPEG frame sync words and the FLAC STREAMINFO block | No duration, no artwork, nothing plays |
| JPG, PNG, GIF | SOI, SOF, DHT and SOS markers; the PNG signature, IHDR and IDAT chunks | Broken-image placeholder in every browser |
| PSD, AI | The 8BPS header, 8BIM resource blocks and the embedded PDF xref | Photoshop: "the file is not compatible" |
| EXE, DLL, APK | The MZ stub, the PE header and the section table; the APK signing block | Windows: "not a valid Win32 application" |
Around 87 extensions have a dedicated profile. Anything else falls back to a generic binary profile that overwrites the header and trailer and scatters entropy through the body, so you can corrupt files online here whether or not the format has a page of its own.
Every file tool says it does not keep your data. Here that claim is checkable in under a minute, three different ways.
Open developer tools, switch to Network, then corrupt a file and sort the requests by size. An upload cannot hide in a list sorted by payload. The only outbound requests belong to the fonts, analytics and advertising tags, and none of them carries your file.
Load this page, turn off Wi-Fi or enable airplane mode, then corrupt a file. It still works, start to finish. A tool that keeps running offline was never sending anything anywhere.
The corrupter ships unminified with no dependencies. In developer tools, open the Sources panel and read it there: under 500 commented lines, every byte operation visible, and not a single fetch, XHR or WebSocket call anywhere in it.
Press the button to generate a 256 KB file in memory, corrupt it, and compare the network requests this page has made before and afterwards. Any third-party analytics or advertising request that fires is named in the result; what matters is that nothing is sent back here, because there is no endpoint to send it to.
Deliberately broken files are a routine part of building, operating and teaching software.
Malformed input is one of the most common causes of production incidents and one of the least exercised paths in a codebase. A corrupted file checks that a parser fails closed rather than hanging or allocating unbounded memory, that an upload endpoint returns a useful error, and that file-type validation reads magic bytes instead of trusting the extension.
A backup you have never restored is a hypothesis. Corrupt a database dump or a deployment artifact, then run the restore runbook against it and find out whether the integrity check fires, whether the import rolls back cleanly, and whether a bad artifact can overwrite a good one.
Generate malformed input for a fuzzing corpus, confirm a scanner handles a broken archive without hanging, or produce reproducible damaged evidence files for digital forensics and incident-response training without touching live malware.
Show a class what a corrupted file looks like at the byte level, demonstrate why checksums and versioned backups exist, and hand students identical damaged files to practise recovery on. The hex view after each run makes the change visible rather than abstract.
Storage fails more often, and more quietly, than most teams plan for. In Understanding the Robustness of SSDs under Power Fault (Zheng et al., USENIX FAST 2013), 13 of the 15 solid-state drives tested across five vendors failed when power was cut mid-write. The observed failure modes were not just dead devices: they included bit corruption, shorn writes, unserializable writes and metadata corruption, which is the category that leaves a file present, the correct size, and unreadable.
Mechanical drives are no safer over time. Google's analysis of more than 100,000 disks reported annualised failure rates climbing toward 8.6% by the third year, with latent sector errors accumulating undetected long before a drive is declared dead.
The practical consequence is that the code handling a damaged file will eventually run in production, whether or not anyone has tested it. Corrupting a file deliberately turns that from an incident into a fixture you can run in CI. For a walkthrough of what to check, see testing the error path nobody tests.
The honest version, including where the alternatives are perfectly fine.
| Capability | Files Corrupter | Typical online corrupter |
|---|---|---|
| Where processing happens | Your browser | Uploaded to a server |
| File size limit | None, to a ~2 GB browser ceiling | Often 10 to 100 MB |
| Corruption method | Format-aware, structural | Random byte flips |
| Defeats Office and Acrobat repair | Yes | Frequently not |
| Damage levels | Four | One |
| Batch processing | Yes | One file at a time |
| Keeps name, size and MIME type | Yes | Usually renamed |
| Works offline | Yes | No |
| Reports what it changed | Profile, markers, hex, SHA-256 | Nothing |
| Account required | Never | Sometimes |
Worth stating plainly, because a tool that overpromises here wastes your time at the moment it matters most.
破損はまず構造に、次にペイロードに適用されます。この順序が「警告付きで開く」と「復元できない」の違いを生みます。
Will not open in any normal application.
| Level | Payload rewritten | Structure targeted | Result |
|---|---|---|---|
| Light | ~0.5% | Magic bytes only | Opens with an error or renders partially. Sometimes recoverable. |
| Standard | ~3% | Header, all signatures, trailer | Will not open in any normal application. |
| Heavy | ~12% | Header, all signatures, double trailer | Defeats recovery and repair software entirely. |
| Shred | 100% | Entire file replaced | Every byte becomes cryptographic noise. |
はい。アカウント、登録、利用制限は一切ありません。ツールはお使いの端末上で動作します。
いいえ。すべてのバイトはブラウザー内でローカルに処理されます。送信も記録も保存も行いません。
Standard 以上では復元できません。復元ツールが頼る構造上の目印を破壊するためです。必ずバックアップを保管してください。
はい。出力は元のファイル名、正確なバイト長、MIME タイプを保ちます。解析しようとしたときにだけ失敗します。
当方からの制限はありません。8 MB 単位で処理するため、数ギガバイトの動画でもメモリーを使い切りません。
すべての形式です。約 87 の拡張子に専用プロファイルがあり、それ以外も汎用バイナリープロファイルで確実に破損します。