Input
- An APP1 XMP packet with sample generator, model, prompt, seed, and workflow values
- An APP2 ICC profile, an EXIF orientation entry, and an APP13 copyright value
- A fixed entropy-coded image scan with known dimensions and a recorded byte hash
Free browser-local tool
Remove AI metadata from images
AI generation metadata can reveal the tool, model, prompt, seed, workflow, or settings used to create an image. This browser-based cleaner inspects supported JPG, PNG, and WebP containers, shows the records it recognizes, and applies the AI metadata preset before downloading a new copy. The default clean keeps dimensions, encoded image data, color profiles, orientation, transparency, animation, and recognized author, copyright, license, attribution, and rights-holder information. Cleaning metadata does not change an invisible signal embedded in the pixels and does not guarantee how any platform will label the image. Both the default clean and a confirmed Full Metadata Reset preserve the encoded pixel payload. The free cleanup runs locally in your browser and does not require an account.
Drop an image to inspect its watermarks
We’ll check for metadata, Content Credentials and supported hidden markers before you choose how to clean it.
Drop, paste or choose an image
JPG, PNG and WebP · up to 20 files · 25MB per file
JPG, PNG and WebP · Only upload files you own or are authorized to edit
Supported removal types
Supported AI metadata removal
The scanner looks inside supported metadata containers rather than guessing from the picture. It can identify supported prompt text, negative prompts, generator and model names, seeds, samplers, workflow labels, software tags, generation parameters, and related XMP, IPTC, EXIF, PNG text, or WebP metadata records. The report separates recognized data from unsupported blocks so the download summary reflects what the preset actually changed.
The AI metadata preset removes its supported AI-generation blocks and a recognized embedded C2PA carrier. It is a fixed preset rather than a field-by-field editor. Content Credentials are still reported separately because provenance data is more than a model-name tag, and recognized rights-management information remains protected in the default mode.
- Supported prompt, model, seed, sampler, and workflow fields
- Generator, editing-software, and generation-parameter records
- Supported XMP, IPTC, EXIF, PNG text, and WebP metadata entries
- A recognized embedded C2PA carrier included by this preset
How it works
How the local cleaner works
The file stays in the current browser session while the tool parses its container and prepares the download.
- 01
Choose or paste a JPG, PNG, or WebP file that you own or are authorized to edit.
- 02
Review the scan by category. Unsupported data is not presented as a verified AI field.
- 03
Run the route's AI metadata preset, or explicitly confirm Full Metadata Reset when you have permission to remove recognized rights and provenance metadata as well.
- 04
Create the cleaned copy. The report lists removed, preserved, and unsupported items and confirms that pixel data was not re-encoded.
- 05
Download the result and, if provenance matters to your workflow, inspect it again with the system that originally created the credentials.
Metadata vs pixels
AI metadata is not an invisible pixel watermark
Metadata is structured information stored around an image's encoded pixels. A prompt in XMP or a generator name in a PNG text chunk can be removed without repainting the scene. C2PA is also file-level provenance data, although its signed structure is more complex than a plain text field.
An invisible pixel watermark is a signal encoded into color or frequency patterns inside the rendered image. Removing a metadata segment does not alter those patterns. This page therefore reports only supported file-container changes; it does not claim to remove SynthID-style signals, defeat a detector, or make content appear human-made.
What is preserved
What the default clean preserves
The normal clean is designed to leave the media payload in place. It retains the image dimensions and supported JPEG scan data, PNG image-data and animation chunks, or WebP image and animation chunks. It also retains recognized ICC color profiles, transparency information, EXIF orientation, and author, copyright, license, attribution, and rights-holder fields.
If removable generation data shares a block with recognized protected rights information, the default clean preserves that block and reports the limitation rather than discard the rights information. Full Metadata Reset may remove a recognized rights or provenance block only after permission is confirmed; it still preserves orientation, ICC color, transparency, animation, dimensions, and the encoded pixel payload.
Limitations
Limits and responsible use
Metadata formats permit private namespaces, compressed packets, duplicated records, malformed lengths, and vendor-specific encodings. The scanner may label such material unsupported or unreadable. Absence of a recognized field is not proof that an image has no AI provenance, and removal of one file-level record does not control copies stored elsewhere or labels applied by a third party.
The tool does not remove visible logos, text overlays, signatures, stock-photo marks, or pixel-level signals. Do not use it to conceal fraud, impersonate another person, remove copyright information without permission, or bypass platform safety and identity checks.
Reproducible example
Test-fixture example: AI fields removed without decoding pixels
A reproducible synthetic JPEG fixture demonstrates container-only cleanup; it contains generated color blocks rather than a user's photograph.
Change
- The AI metadata preset omits the dedicated AI XMP packet from the rebuilt JPEG container.
- The ICC, orientation, and copyright records are copied to the output.
Verification
- The output opens as a JPEG with the same width and height.
- The entropy-coded scan bytes have the same hash before and after, so this fixture's pixels were not re-encoded.
- A second scan reports the supported AI block absent and the preserved records present.
FAQ
Questions about this tool
Does this remove every sign that an image was made with AI?
No. It removes only the supported file-level records included in this route's preset. It does not alter invisible pixel signals or information held by a platform, publisher, or earlier copy.
Will the image look different?
No supported cleanup mode decodes or re-encodes the image payload. The default preset and a confirmed Full Metadata Reset both preserve the pixel stream, dimensions, orientation, ICC color, transparency, and animation.
Why are copyright and author fields still present?
They are rights-management information, so the default preset intentionally keeps them. Remove rights information only from content you control and only through an explicit, confirmed reset.
Is my prompt uploaded?
No prompt text or image content is sent to run the free browser-local scan and clean. Avoid sharing the report if it contains information you consider sensitive.
Related tools and guides
Continue with a verified path
Ownership notice
Only process images you own or are authorized to edit. Do not remove provenance or rights information to misrepresent authorship, licensing, identity, or the history of third-party content.