Remove the AI label. Keep every pixel.
Strip C2PA Content Credentials, XMP, EXIF and video tags from JPG, PNG, WebP, MP4 and MOV files. Nothing is re-compressed, there is no file size limit, and nothing leaves your browser.
- C2PAContent Credentials manifest · marked AI-generated
- XMPreferences Content Credentials
- Video tagsmoov/udta user data
- Tool nameshandler and encoder names
- Keptvideo and audio streams, byte-for-byte
What gets removed
An "AI info" or "Made with AI" label usually starts with data hidden inside the file, not with anything you can see in the picture. These are the places that data lives, and all of them are cleared.
Content Credentials
A signed manifest that records which app created or edited the file, often with a flag saying the content was AI-generated. Stored in JPEG APP11 segments, a PNG caBX chunk, a WebP C2PA chunk, or an MP4/MOV uuid box.
AI source fields
The IPTC DigitalSourceType field can say trainedAlgorithmicMedia (AI-generated) or compositeWithTrainedAlgorithmicMedia (AI-edited). XMP also stores the creator tool, such as an editor or generator name.
Camera, software & GPS
Camera make and model, the software that last saved the file, timestamps and GPS coordinates. Useful to you, but nobody else needs them.
Prompts & generation settings
Stable Diffusion, ComfyUI and similar tools write the full prompt, seed, sampler and model name into PNG text chunks. Anyone can read them with a basic viewer.
MP4 & MOV tags
User-data and metadata boxes (location, device, title, comments), embedded XMP, C2PA boxes, and the handler and encoder names that reveal which app exported the clip.
What we deliberately keep
The image or video data itself, the ICC color profile (so colors don't shift) and the photo's orientation (so phone shots don't end up sideways).
Why "lossless" matters
There are two ways to get rid of metadata. The quick way is to draw the image onto a canvas and export a brand-new file. Many online removers do exactly that. It works, but it costs you: JPEGs are compressed a second time, WebP and AVIF often come back as much larger PNGs, the color profile is dropped (wide-gamut phone photos can look washed out), and some tools add noise or resize on purpose.
This tool takes the other route. It reads the file's structure, cuts out the metadata segments and stitches the rest back together. The compressed image data is copied byte-for-byte, so the result decodes to exactly the same pixels as the original. For video, metadata boxes are blanked in place, which keeps every offset valid, so a 2 GB clip is processed in a few seconds without being loaded into memory.
| Redraw on canvas | Byte-level removal (this tool) | |
|---|---|---|
| Image quality | Re-compressed, small loss each time | Identical pixels |
| File format | Often converted to JPG or PNG | Same format as you uploaded |
| Color profile | Usually dropped | Kept |
| Video | Needs a full re-encode or a server | Blanked in place, no re-encode |
| Large files | Limited by memory, often 10 to 15 MB caps | No size limit |
How it works
- Drop your filesPhotos and videos, as many as you like. They are opened by your browser, not sent anywhere.
- See what's insideEach file gets a report: Content Credentials, AI source fields, camera and GPS data, prompts, video tags.
- Removed and re-checkedThe cleaned file is parsed again from scratch. You only see "clean" if that second pass finds nothing.
- DownloadOne file at a time or everything in a single .zip. The original on your device is never modified.
Will this remove the AI label on Instagram, TikTok or LinkedIn?
Often, yes, but not always, and it is worth knowing why before you rely on it.
Instagram, Facebook and Threads. Meta says it labels images by reading the "AI generated" information in the C2PA and IPTC standards, and names Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock as the tools it can recognise (Meta, Feb 2024, updated Apr 2025). Those are exactly the fields this tool removes. Meta also asks people to self-disclose photorealistic AI video and audio and says it may apply penalties when they don't. In 2024 it renamed "Made with AI" to "AI info" after photographers complained that real photos edited with AI-powered tools were being labeled (TechCrunch), and later moved the label into the post menu for edited content (TechCrunch).
TikTok. TikTok reads C2PA Content Credentials to automatically label AI-generated images and videos uploaded from other platforms (TikTok newsroom, May 2024). A file without a manifest gives it nothing to read.
The limit: watermarks inside the pixels. OpenAI now adds both C2PA metadata and a SynthID watermark to images made with ChatGPT and its API, and notes that metadata "can easily be removed either accidentally or intentionally" (OpenAI Help Center). Google embeds SynthID in content from Gemini, Imagen, Veo and Lyria (Google, May 2025). A watermark woven into the image is not metadata, so this tool cannot remove it, and whether a given platform checks for it is that platform's decision. Platforms can also apply labels when content is self-declared. More on SynthID.
So the honest summary: removing metadata clears the file-level signals platforms say they read. It does not change what the image is, and it cannot remove a pixel watermark.
Who this is for
Photographers whose edits were tagged as AI
In 2024 photographers complained that real photos edited with AI-powered tools were being labeled "Made with AI", and Meta renamed the label as a result (TechCrunch). If a photo is genuinely yours, you have disclosed AI use where it matters, and you don't want an automated label on it, this clears the file-level signal.
Anyone sharing photos privately
The same files that carry AI flags also carry GPS coordinates, device serial numbers and timestamps. Stripping them before you post or send a file is basic privacy hygiene.
Creators and teams with big batches
No file count limit, no size cap and no account. Drop a folder's worth of exports, then download one .zip.
How we test it
Every release is run against the official sample files published by the C2PA project, plus files we sign ourselves with a test certificate and an "AI-generated" claim, in every supported format. A file only passes if three independent checks agree:
- The official C2PA library can no longer find a manifest in the cleaned file.
- A raw byte scan finds none of the known markers: JUMBF and C2PA boxes, XMP packets, IPTC blocks, generator names, prompts or GPS tags.
- The content is identical: images decode to exactly the same pixels, and every video and audio packet has the same checksum as the original.
On a 760 MB MP4 the analysis step took under 10 milliseconds and used less than 1 MB of extra memory, because the video data is never read into memory at all.
Frequently asked questions
Is my file uploaded anywhere?
No. Everything runs in your browser using JavaScript. The page makes no network requests when you process a file, and it keeps working if you go offline after it has loaded.
Does this change the image or video quality?
No. Only metadata segments are removed. Image data and video and audio packets are copied byte-for-byte, so the cleaned file decodes to identical pixels and frames.
Which formats are supported?
JPG/JPEG, PNG, WebP, MP4 and MOV (including M4V). AVIF, HEIC and GIF support is in progress.
Is there a file size or batch limit?
No. Videos are processed in slices, so even multi-gigabyte files work. Batches download as a single .zip up to about 4 GB in total; beyond that, files download one by one.
Can it remove SynthID or other invisible watermarks?
No. SynthID and similar watermarks are woven into the pixels themselves, not stored as metadata, so no metadata tool can remove them. Here is how SynthID works.
Why is my phone photo still upright after cleaning?
Orientation is stored in EXIF. When a photo needs rotating, we keep a minimal EXIF block that holds only the orientation value and nothing else.
Is it free?
Yes, free with no account and no daily limit.