Image Metadata & AI Provenance Cleaner

Image metadata is information stored inside an image file next to the picture. Some of it describes the picture, such as its size and colour profile. Some describes how it was made, such as the camera or app. Some describes you, such as an author name, a location or a caption you typed once and forgot.

It is not secret or sinister, and most of it is harmless. It is worth knowing about because it travels with the file when you send it. A photo you share to show a view may also carry the exact spot you stood.

EXIF is the block that cameras and phones fill in automatically. A typical photo has Make and Model, DateTimeOriginal, exposure fields such as ExposureTime, FNumber and ISO, an Orientation value, and often GPS tags. Some cameras also write body and lens serial numbers, and a vendor-specific “MakerNote” that this tool reports but does not decode.

Not every device writes every field, and edited files can hold values from several programs. Treat the fields as what the file claims, not as verified facts.

Phones write the position to help organise photos on a map. The tags hold latitude, longitude and often altitude, direction and a GPS timestamp. A pair of coordinates can be accurate to a few metres.

Example: a photo taken in a garden at home carries the coordinates of that garden. Posted with its metadata intact, it tells anyone who reads the file where you live. This tool shows these fields hidden by default, so you can look without exposing them on a shared screen, and it never opens a map or sends them anywhere.

1. Choose a file, or drag it onto the upload area. Nothing is uploaded.
2. Open Metadata and Privacy Findings to see what the file carries.
3. Open Clean & Export, pick a mode and read the “what will happen” summary.
4. Choose Clean Metadata, review the verification, then download the cleaned copy.

Your original file is never changed. The cleaned copy is named with -cleaned and its real extension.

For JPEG, PNG, WebP and GIF the tool reads the file the way a decoder does, by its real structure: JPEG segments, PNG chunks, WebP RIFF chunks and GIF blocks. It never searches for byte patterns. Each metadata block is kept, removed or rewritten, and every other byte is copied across unchanged. Container sizes, PNG checksums and WebP flags are rewritten so the result stays valid.

EXIF is rewritten field by field when you keep some fields and remove others. XMP is edited by cutting the exact properties out and re-reading the result, and if that cannot be validated the whole packet is removed rather than leaving a half-edited one. A file that looks damaged is not changed.

The fourth mode, Re-encode, draws the picture onto a canvas in your browser and saves it again. It is not lossless, it can change colours and size, and it is not offered for animated images.

After cleaning, the tool reads the new file from scratch with the same parser and compares it with the original. It checks that every field you chose to remove is gone, that the file is structurally valid, that dimensions, animation, transparency and orientation are as expected, and that the compressed image data is identical. It then reports what was removed, what was kept on purpose, and what it could not check.

The result is worded “verified against supported metadata checks” because it only covers structures this tool reads. For a second opinion, load the cleaned file back in or open it in a separate metadata viewer.

IPTC is a standard from the news industry for captions, headlines, keywords, creator and credit lines, copyright notices, contact details and place names. XMP is an XML format from Adobe that can hold many field types, including titles, ratings, editing history and creation tools, and newer labels such as a digital source type.

Both can duplicate what is in EXIF. A photo may show its author in all three places, so removing only one leaves the name in the others. This tool reads and removes all three.

Content Credentials are a signed provenance record, defined by the C2PA standard, that can be embedded in an image. A record can say which software or device created the file and what edits followed, including whether generative AI was used. It is stored in a dedicated container in the file, not in EXIF or XMP.

This tool finds the container in JPEG, PNG and WebP (and AVIF or HEIC for inspection), reads its structure and lists the claimed actions. It does not check the signature, the certificate chain, or that the record belongs to this image. A record that reads cleanly can still be forged or stale, so use a dedicated verifier if validity matters.

No. Metadata and watermarks are different things. Metadata is information stored beside the pixels. An invisible watermark or statistical signal is embedded in the pixels themselves, and it can survive metadata removal, resizing and re-encoding. This tool does not detect such signals and does not claim to remove them.

It also is not an AI image detector. It can show you AI-related metadata that a file declares, such as a digital source type label or a generator’s text in a PNG. A file without these is not shown to be human-made, and one with them is not proven to be generated.

Not with the three modes that edit the file’s structure. The compressed image data is copied without being decoded, so quality, colours and pixels are untouched, and the tool compares a fingerprint of that data before and after to prove it. Files usually get a little smaller.

Re-encoding is different. Decoding and compressing again can add artefacts, change colours and alter the size, which is why it is a separate, clearly labelled mode and never described as lossless.

Location, device serial numbers, author and contact fields, timestamps, descriptions, thumbnails and editing history are generally safe to remove before sharing. Be more careful with three things:

• Orientation tells viewers to rotate the picture. Removing it can make a photo appear sideways, so this tool keeps it.
• Colour profiles describe how to display colours. Removing them can shift colours in wide-gamut images.
• Copyright and credits are your attribution. Removing them removes it from the file, though not from your rights.

AI labels and Content Credentials are disclosure. Some platforms and rules expect them to stay, so this tool keeps them in the default mode and only removes them if you choose to.

Usually not. A screenshot is a new image created by your device, so it does not normally copy the EXIF or GPS tags of the photo it shows. It has its own metadata instead, which varies by device and app and can include the device model and the time. The pixels still show everything visible, such as faces, names and signs. If a screenshot matters for privacy, inspect it like any other file.

Re-encoding in a browser drops most embedded metadata blocks, but not everything that identifies an image. Pixel-level watermarks can persist. The picture still shows what it shows. Files can be named after a place or person. And a browser may add its own small header. After a re-encode, inspect the result: this tool does that automatically and lists what it cannot check.

Platform behaviour also varies. Many sites rewrite or strip some metadata on upload, but which fields, and when, differs between services, apps and versions, and the service may still have received the original. Do not rely on a platform to clean your files for you.

Remove location and device details first, then keep or remove author and copyright depending on whether you want attribution. Use Remove metadata for everyday sharing, Maximum cleanup if you also want AI labels and credentials gone, and Selective cleanup if you want to keep your copyright line but drop the location.

Check the Before & After tab, download the cleaned copy and upload that one. If you work with logs or tokens as well, the Log Sanitizer and JWT Decoder follow the same local-first approach.

The file is read with your browser’s own file API and processed by code running in this tab. This tool does not send the image, its metadata, its file name, coordinates, a hash or the report to any server, and it does not store them in local storage, session storage, IndexedDB or the cache. It never fetches addresses found inside an image, such as a manifest URL in XMP.

Two honest limits. First, this page, like every page on the site, loads Google Tag analytics and Google AdSense. They run outside this tool and see ordinary page-visit information, and this tool does not hand them your image or its metadata. Second, the browser keeps the file in memory while you work. Reset releases it, but browsers do not guarantee memory is wiped. For highly sensitive material, work on a device and browser profile you trust.

Related reading: the HTTP Security Headers Analyzer and CSP Analyzer show how a site’s own headers limit what scripts can do.

• Not supported: SVG (it can contain scripts and external references, so it is never opened), JPEG XL and camera RAW formats.
• Limits: 50 MB per file, 120 megapixels, and 48 megapixels for re-encoding.
• One image at a time. There is no batch mode in this release.
• Re-encoding is not offered for animated images, which it would flatten.
• MakerNote and other proprietary blocks are reported but not decoded. Data stored after the image is reported but not read inside.
• Invisible watermarks, pixel-level signals and C2PA signature validity are not checked.
• This release has no worker thread, so very large files can briefly pause the page while they are read.