AI Label RemoverRemove AI Detection From Image: Strip the C2PA & XMP Records Detectors Read
Remove the AI provenance records hidden in your image before you upload it — right here in the browser, nothing uploaded. Watermarks inside the pixels, such as SynthID, are a different thing and stay.
- Works in the browser · no app, no signup
- Free · nothing uploaded · no watermark
- JPG · PNG · WebP · HEIC · AVIF · 100 images at once
What a detector actually reads
“AI detection” can mean two different checks. One reads the records stored in the file; the other judges the picture itself. Removing AI detection from an image only works for the first.
Records in the file
This tool removes them
- C2PA / Content Credentials — a signed record saying the image was generated or edited with AI
- XMP and IPTC fields such as the digital source type and the AI tool’s name
- EXIF tags like Software or Credit naming an AI app, plus GPS and camera details
- AI prompts and settings written into PNG text chunks
Signals in the pixels
No metadata tool can remove them
- Invisible watermarks such as SynthID, woven into the image itself
- Visible badges, like the “AI-generated content” mark some phones add
- Classifier scores — a model judging whether the picture looks AI-made
How to remove AI detection from an image
Open the original file
Load the image straight from where it was saved. Re-saving it through a chat app or messenger first can strip or rewrite the very records you're trying to inspect.
See what the detector reads
The image is cleaned as soon as it loads, and every C2PA manifest, XMP packet and EXIF block found in the file is listed, with the ones that mark it as AI-generated called out.
Strip the provenance records
Keep All metadata to remove every block, or pick AI markers only to remove just the C2PA record and AI fields and keep camera details. The image pixels are never re-encoded or altered.
Download the verified copy
Each cleaned file is read again automatically before you save it — no AI markers left, image data byte-for-byte identical. Upload the cleaned copy, not the original.
What gets stripped
Three records carry most AI detection signals in an image file. Here is what each one says and what happens to it; the colour profile and orientation always stay, so the image looks the same.
| Record | What it says | When you clean it |
|---|---|---|
| C2PA manifest | A signed Content Credentials record: which app or AI model made or edited the image, and when. Adobe, OpenAI, Google, Microsoft and Samsung tools write one. | Removed in both modes. This is the record platforms rely on most. |
| XMP packet | Edit history and AI fields such as the IPTC digital source type (for example “trainedAlgorithmicMedia”) and the name of the AI tool. | Removed. AI markers only blanks just the AI fields and keeps the rest. |
| EXIF block | Camera, date and GPS, plus text tags like Software or Credit that can name an AI app. | Removed. AI markers only clears only the tags that name AI, and GPS unless you keep it. |
Free, no upload, no account
You can remove AI detection from image files free of charge, with no account and no watermark added. The cleaning runs inside your browser, so the image never leaves your device — up to 100 files at a time, 50 MB each, downloaded one by one or as a ZIP.
By file type and device
These are the common places AI records get written into an image — at capture, after an AI edit, or on export.
- iPhone (HEIC) — HEIC photos are cleaned as HEIC, with no conversion. Since iOS 27.2, images made or edited with Apple Intelligence can carry a SynthID watermark in the pixels — removing metadata leaves that in place.
- Android — Pixel 10’s camera adds Content Credentials to every JPEG it takes, and Google Photos adds them to images edited with its AI tools. Samsung Galaxy AI edits such as Generative Edit get a C2PA record and a visible “AI-generated content” mark in the corner; the record can be removed, the visible mark can’t.
- Windows — Images generated with Cocreator in Paint contain a C2PA manifest. Clean the exported file before you upload it.
- Mac — From macOS 27.2, most images created or edited with Apple Intelligence include a SynthID watermark. Photoshop adds Content Credentials when you use its generative AI features, on Mac and Windows alike.
- Any device — Exporting from an editor writes metadata again. Clean the file you are actually going to upload, after the last edit, not the original from earlier.
Why your image still gets flagged
- The platform checks the pixels — Some platforms and standalone detectors run their own analysis after upload. Cleaning the file removes what it says about itself, not what the picture looks like.
- A watermark is in the image data — Images from Google’s AI tools, and Apple Intelligence images from iOS 27.2, carry SynthID in the pixels. It survives cleaning by design.
- The post is already published — A label attached to a live post stays there. Delete the post and upload the cleaned copy instead.
More on invisible watermarks like SynthID and what they mean for detection.
Frequently asked questions
What does "AI detection" actually look at in an image?
Two different things. Social platforms read the file’s metadata first — a C2PA manifest or XMP field recording that the image was generated or edited by AI. Many standalone detectors score the pixels themselves instead, and some look for invisible watermarks such as SynthID. This tool removes the metadata layer only.
Can this make a detector stop flagging my image?
It removes the metadata that platforms and metadata checkers read, which is the common reason an image gets labelled. It cannot change pixel-level signals: if a detector is scoring the image content or reading an invisible watermark, cleaning the file will not change its verdict.
Does it change the image itself?
No. Only the metadata blocks are removed and the image data is never re-encoded, so resolution, colour and sharpness stay identical. Each cleaned copy is compared byte for byte with the original before you download it.
Is there a free version?
Yes, removing AI detection metadata from an image is free in the browser — no account, no watermark and no upload. You can clean up to 100 files at a time, 50 MB each.
Which file types work?
JPG, PNG, WebP, AVIF and HEIC/HEIF straight from an iPhone. There is no need to convert to JPG first, which matters because converting can add its own metadata on export.
Can it remove AI detection from video?
Not yet — this tool handles images only. AI provenance records in video files work differently, and video support is planned separately.
Why is my image still labelled after cleaning?
Three common reasons: the platform analyses the pixels after upload, an invisible watermark such as SynthID is embedded in the image data, or the label was attached to a post that is already published — editing the file cannot change a label already on a live post.
Is removing these records legal?
Removing metadata from files you own is legal in most countries and is routine for privacy. It is not a way to pass AI images off as real photographs: platform rules can still require you to disclose AI content, and laws such as the EU AI Act require disclosure of deepfakes. Only clean files you own or have permission to edit.