Remove the AI label from your photos before you post them
Facebook, Instagram and Threads add an "AI Content" label when a photo carries certain hidden tags. TikTok does something similar with its AI-generated content label. Those tags live inside the file, not in the picture you see.
This AI label remover clears them. It strips the C2PA Content Credentials, the IPTC and XMP "made with AI" fields and every other piece of metadata, and it leaves the picture exactly as it was.
Why does Facebook say "AI Content" on my photo?
Meta reads standard signals inside the image file. The two main ones are a C2PA manifest, which tools like ChatGPT, Adobe Firefly and Gemini sign into the file, and the IPTC field digitalSourceType set to trainedAlgorithmicMedia. When Facebook finds one of them, the post gets the label.
That is also why real photos get labelled. Edit a picture with an AI feature such as Generative Fill, even on a small corner, and the saved file can carry Content Credentials that describe the whole image as AI-edited.
How do I remove the AI label from a Facebook post?
Fix the file before you upload it. Facebook reads the tags at upload time, so the way to post without the label is to post a cleaned copy.
- Upload your photo on this page.
- Download the cleaned copy.
- Post the cleaned copy on Facebook.
A label that is already on a live post stays with that post. Delete the post and upload the clean version instead.
Does it work for Instagram, Threads and TikTok?
Yes. Instagram and Threads use the same Meta labelling system as Facebook and read the same tags. TikTok reads C2PA Content Credentials to label AI-generated content. In each case the label starts with data in the file, and the cleaned file no longer carries it.
Images from ChatGPT, Gemini and Adobe Firefly
ChatGPT and DALL-E images arrive with a signed C2PA manifest. Gemini and Imagen images can carry Google's own provenance tags. Adobe Firefly signs Content Credentials into everything it produces. The report tells you which of them it found before it removes the marks.
When the AI label is a false positive
Many labelled photos were taken with an ordinary phone or camera. The label appeared because one app in the editing chain wrote an AI tag, sometimes for a small fix like erasing a stray object. Clearing the metadata removes that tag together with the rest of the editing history.
What this AI tag remover takes out
AI tags are the main job. The same pass also clears everything else the file knows about you and your device.
- C2PA and Content Credentials manifests
- IPTC and XMP fields such as digitalSourceType and trainedAlgorithmicMedia
- "Generated by" strings in PDF and Office documents
- Invisible zero-width characters in AI-written text
- Author names, GPS location, camera make, model and serial number
- Creation dates, modification dates and editing history
Does it tell you which AI made the file?
Yes. The file is inspected before anything is removed, so the report names the generator instead of only saying something was removed. It recognises OpenAI (ChatGPT, GPT, DALL-E), Google (Gemini, Imagen), Anthropic (Claude), Adobe Firefly, Midjourney, Stability AI, xAI (Grok), Leonardo AI, Microsoft Copilot and Black Forest Labs (FLUX).
A file whose tags name none of them is still fully cleaned. The report just says the generator was not named in the file.
What if my file had no AI tags at all?
Then the report says so. A file with nothing hidden is marked Verified clean and reads "No AI provenance marks were found in this file." The file comes back unchanged because there was nothing to take out.
That answer is useful too. If a photo with no AI tags still gets labelled, the platform used a different signal, such as your own AI disclosure when posting, and you know that before you upload.
Will my photo lose quality?
No. A standalone image is never decoded or re-compressed. Your PNG comes back with byte-identical pixel data and your JPEG keeps its scan data untouched, so sharpness and detail stay the same. The colour profile is removed along with the rest of the metadata, so a photo saved in a wide-gamut profile such as Display P3 can look slightly different; a standard sRGB image looks the same.
There is one exception. An AI mark inside a picture that is embedded in a PDF can only be reached by rewriting that picture, so its pixels change very slightly. The report counts and names every image this happened to.
Can it find AI tags inside a PDF's pictures?
Yes. A scanned page or an exported report can hide a C2PA manifest inside the embedded picture instead of the document properties. A cleaner that only clears the document layer misses it. This tool opens the embedded pictures and cleans them as well.
Does it remove a visible AI watermark?
No. A logo, a caption or a mark printed into the pixels is part of the picture, and removing it means repainting the image. That is a job for an image editor. This tool works on the hidden layer: tags, signed manifests and invisible characters.
The invisible characters in AI-written text
Some AI models leave zero-width spaces and joiners in the text they write. They are invisible on screen and survive copy and paste into a document, an email or a CMS. This tool finds them and removes them.
Plain-text files get two extra options you can switch on: normalising look-alike Unicode characters and mapping confusable letters back to plain ASCII. Both change the text itself, so they apply only to text files.
Which files does it work on?
The type is read from the file's own bytes, so a file saved with the wrong extension still comes back in the right format. Original filenames are kept, Arabic and Chinese names included. For a password-protected PDF, enter the password and the cleaned copy comes out unlocked.
| Category | Formats |
|---|---|
| Images | PNG, JPG / JPEG, WebP, AVIF, HEIC / HEIF, GIF, BMP, TIFF, SVG |
| Documents | |
| Office | DOCX, XLSX, PPTX |
| Books | EPUB |
| Text | HTML, Markdown, TXT |
How do I know it worked?
The tool opens its own output a second time and checks for the C2PA manifest and every AI tag it removed. A file is reported clean only when they are gone. If a file's structure resisted a full rewrite, the report says so instead of claiming success.
The removal engine is the open-source, MIT-licensed watermarks-remover project, so the code that touches your files is public.
What you get
- The hidden AI tags that trigger the AI Content label removed before you post
- C2PA manifests, IPTC and XMP AI fields and "Generated by" strings cleared from images, PDFs and Office files
- A report that names the AI behind each file, or marks it Verified clean
- Standalone images returned byte-identical, with no quality loss
- GPS, camera, author and editing history cleared in the same pass
- Output checked a second time, so clean means confirmed
What's next?
Need only the camera and location data off a photo? Remove Photo Metadata does exactly that. Cleaning a document instead? Remove PDF Metadata clears the author, producer and dates, and Flatten PDF makes filled fields and annotations permanent before you send it.