Remove AI label metadata from photos for free: find and strip C2PA, XMP, IPTC and EXIF AI fields in your browser. Pixels and invisible watermarks stay.
Removes metadata only. Invisible watermarks (such as Google SynthID) and visual AI detection are not affected.
Three files from AI image models and one camera photo, with this tool’s real results.
Found
AI Label RemoverC2PA generator OpenAI Media Service APIC2PA software agent gpt-imageC2PA digitalSourceType trainedAlgorithmicMediaAfter “Remove all metadata”
Found
C2PA action Created by Google Generative AI.C2PA action Applied imperceptible SynthID watermark.C2PA digitalSourceType trainedAlgorithmicMediaAfter “Remove all metadata”
Found
C2PA software agent Volcengine_Ark_CNC2PA model doubao-seedream-5-0-proC2PA digitalSourceType trainedAlgorithmicMediaAfter “Remove all metadata”
No AI metadata
Only some AI labels come from the file. A tool to remove AI label data helps with the first three rows.
| Label source | What it looks like | Removable here? | What to do |
|---|---|---|---|
| Content Credentials (C2PA) | A signed manifest, e.g. generator "OpenAI Media Service API" or the action "Created by Google Generative AI." | Yes | Clean it. Both modes remove C2PA. |
| IPTC digital source type | DigitalSourceType = trainedAlgorithmicMedia or compositeWithTrainedAlgorithmicMedia, in XMP or IPTC | Yes | Clean it. AI-only mode removes just this field and keeps the rest. |
| Generator names and prompts | EXIF Software, XMP CreatorTool, PNG text with a prompt or a ComfyUI workflow | Yes | Clean it. |
| Invisible watermark | A pattern inside the pixels, such as Google SynthID. Not visible in any metadata viewer. | No | No metadata tool can remove it. |
| Visual AI detection | The platform looks at the image itself. Nothing in the file. | No | If the photo is real, use the platform’s appeal. |
| Your own disclosure | You switched on an AI label when posting, or the platform requires one. | No | Turn the toggle off if the content allows it. Required labels should stay. |
A photo you took yourself can still carry an AI marker. Editors with a generative feature (AI fill, expand, object removal) can record it as a digital source type like compositeWithTrainedAlgorithmicMedia or in a C2PA manifest. Platforms read that flag, not the camera data.
To remove AI metadata from a real photo and keep its camera data, scan it and choose Remove AI fields only. If the verdict is “No AI metadata”, the label does not come from the file; appeal it on the platform.
The default. Removes every EXIF, XMP, IPTC, C2PA and text block, keeping only orientation and the color profile.
Removes C2PA and only the fields that point to AI. Camera data stays exactly as it was.
The compressed image data is copied byte for byte, never decoded and saved again, so the file keeps its format.
| Option | Best when | Not ideal when |
|---|---|---|
| This browser tool | You want to see the AI field first, keep full quality and clean a batch without installing anything. | The label comes from pixels, not metadata. |
| Command-line tools such as ExifTool | You use the terminal and want scripts. | You need to know which field triggers the label. |
| Screenshot or re-export | Quality does not matter. | You care about resolution or colors. Pixel watermarks come along too. |
Meta says it labels images from industry-standard indicators and from what people disclose, and that content only edited with AI tools shows its “AI info” label in the post menu (Meta’s labeling approach). TikTok says it may label a post automatically when the upload carries C2PA Content Credentials, that this auto label can’t be removed from the post, and that creators must label realistic AI-generated content (TikTok: About AI-generated content).
Drop the photo into the tool above. If it shows AI metadata (C2PA, a digital source type or a generator name), clean it and upload the cleaned copy. If it finds none, the label comes from a pixel watermark, visual detection or your own disclosure, and cleaning will not change it.
Platforms read standard fields in the file. If an AI generator or a generative edit wrote them, they are there on every upload. Remove AI metadata once and upload the cleaned copy.
Yes, if the hidden data is metadata. This tool removes C2PA manifests and the XMP, IPTC, EXIF and PNG text fields that name an AI source. It cannot remove a watermark that is part of the pixels.
Only when the label is triggered by metadata. Remove AI metadata and the signal is gone from the file; a platform may still label the post from its other checks.
Not with a file tool: the platform keeps the copy you uploaded. Clean the original, then delete the post and upload the cleaned image.
It is free and needs no account. Everything runs in your browser, so there is nothing to sign up for.
No. The tool rewrites only the metadata blocks and copies the compressed image data unchanged, so the cleaned file decodes to exactly the same pixels.
Yes. We tested it on real files from ChatGPT and gpt-image, Google Gemini (Nano Banana) and ByteDance Seedream: it found their C2PA data and removed it. For other generators it checks the same standard fields and known generator names.
No. SynthID is embedded in the pixels, and this tool never changes pixels. Google’s C2PA data records it: “Applied imperceptible SynthID watermark.”
Yes. HEIC files are cleaned without converting them to JPG, and the photo keeps its rotation.
If the label comes from metadata, cleaning removes the signal. If the platform requires you to disclose AI content, keep the label: removing metadata does not change that rule.
See what the file says about AI, then remove AI label data in one click. Nothing leaves your device.
Remove AI label now
