Free · runs in your browser
AI Metadata Remover
Remove AI metadata from photos and videos field by field. The AI metadata remover shows the C2PA manifest, the IPTC source type and the XMP and EXIF tags inside your file, takes them out and leaves the image data untouched.
- Files never leave your device
- Every AI field listed
- Camera details kept
- Photos and videos
Inside the file
- IHDR · size
- zTXt · IPTC credit
- iTXt · XMP AI fields
- caBX · C2PA (Google)
- IDAT · image data
ai-vase.png
INSIDE THE FILE
Where AI metadata sits in each file format
Every format keeps AI metadata in its own blocks. Pick a format to see where the AI metadata remover finds it, what comes out and what stays.
APP1 · EXIF
Camera, lens, date taken, GPS, orientation
Kept
APP1 · XMP
Iptc4xmpExt:DigitalSourceType, photoshop:Credit “Made with Google AI”, the AI app as CreatorTool
AI fields removed
APP2 · ICC
Color profile
Kept
APP11 · C2PA
Signed Content Credentials: the app, each edit, trainedAlgorithmicMedia
Removed
APP13 · IPTC
Caption, credit and copyright lines in one block
Removed when it holds an AI credit
Image data
The compressed picture
Copied byte for byte
A JPG is a row of segments before the picture. The AI metadata remover drops the C2PA segment, takes the AI fields out of XMP and IPTC and copies every other segment, and the picture, exactly as it was.
IHDR
Width, height and color type
Kept
zTXt · IPTC
An IPTC block stored as hex text; Google writes “Made with Google AI” here
Removed
iTXt · XMP
DigitalSourceType and the AI credit line
AI fields removed
tEXt · parameters
Prompt, seed and model from Stable Diffusion or ComfyUI
Removed
caBX · C2PA
Signed Content Credentials from ChatGPT, Gemini or Firefly
Removed
IDAT · image data
The compressed picture
Copied byte for byte
PNG files from AI apps can carry several kinds of AI metadata at once: a C2PA chunk, XMP, an IPTC block written as hex text and, from Stable Diffusion or ComfyUI, the whole prompt. Each chunk has its own checksum, which the AI metadata remover keeps valid.
VP8X
Which extra chunks the file has
Updated to match
VP8 · image data
The compressed picture
Copied byte for byte
EXIF
Camera details, or an AI app named as Software
Only the AI app name goes
XMP
DigitalSourceType and AI credit lines
AI fields removed
C2PA
Signed Content Credentials
Removed
WebP keeps metadata in chunks next to the picture and flags them in its VP8X header. The AI metadata remover updates those flags and the file size so the WebP still opens everywhere.
ftyp
File type
Kept
uuid · C2PA
Content Credentials, for example from Adobe Firefly
Removed
uuid · XMP
XMP with DigitalSourceType
AI fields removed
moov · udta
Device, GPS and tags, AI tags such as the generator name
AI entries removed; a box that mixes AI text with other tags goes whole
mdat · frames
The video and audio
Not re-encoded
Videos are not re-encoded: the boxes around the frames are rebuilt and every frame stays where it was.
BY APP
The exact AI metadata fields in ChatGPT, Gemini and Photoshop files
Every AI app writes its own mix of AI metadata. These are the fields the AI metadata remover finds in files from the most common ones, read from real files or taken from the maker’s own documentation.
ChatGPT
OpenAI- C2PA claim generator
- = OpenAI Media Service API
- c2pa.created
- = trainedAlgorithmicMedia
- c2pa.watermarked.unbound
- = a mark in the pixels too
- signed by
- = OpenAI OpCo, LLC
Read from a GPT Image file on 21 Oct 2026
Gemini and Nano Banana
Google- C2PA signed by
- = Google Media Processing Services
- Iptc4xmpExt:DigitalSourceType
- = trainedAlgorithmicMedia
- photoshop:Credit
- = Made with Google AI
- IPTC Credit (PNG zTXt)
- = Made with Google AI
Read from a Nano Banana file on 21 Oct 2026
Google Photos
Magic Editor, Magic Eraser, Zoom Enhance- Iptc4xmpExt:DigitalSourceType
- = compositeWithTrainedAlgorithmicMedia
- Credit
- = Edited with Google AI
Google, announced through IPTC in October 2024
Photoshop and Firefly
Adobe- C2PA actions
- = c2pa.created, c2pa.edited
- digitalSourceType
- = trainedAlgorithmicMedia or compositeWithTrainedAlgorithmicMedia
Adobe’s Content Credentials documentation
Stable Diffusion
AUTOMATIC1111 web UI- PNG tEXt parameters
- = prompt, Negative prompt, Steps, Sampler, Seed, Model
The web UI’s own PNG info format
ComfyUI
Node-based Stable Diffusion- PNG tEXt prompt
- = the node graph as JSON
- PNG tEXt workflow
- = the full workflow
ComfyUI’s saved-image format
CHECK FIRST
How to check a photo for AI metadata
The AI metadata remover is also an AI metadata checker: it lists what it finds before anything is removed.
- 1
Drop the file into the AI metadata remover above
It reads every block of the file on your device, so nothing is uploaded, and works on JPG, PNG, WebP, HEIC, AVIF, MP4 and MOV.
- 2
Read the report
Each AI field gets a row with where it is stored, the app that wrote it and the platforms that read it. Camera, location and other details are listed apart, marked Kept.
- 3
Compare with a C2PA check if you like
Adobe’s Content Credentials Verify site shows the C2PA manifest on its own. The AI metadata remover also reads the IPTC, XMP and EXIF fields and the PNG prompts that a C2PA check skips.
LINE BY LINE
What the AI metadata remover takes out of the sample
The metadata of the vase image above, read from the file. Struck-out lines are the AI fields that come out; the rest stays.
BEYOND METADATA
What an AI metadata remover cannot reach
The AI metadata remover clears every AI field in the file. These marks live outside the metadata, so no metadata tool touches them.
Pixel watermarks
Google SynthID and similar watermarks are woven into the pixels. The AI metadata remover copies the pixels of JPG, PNG, WebP and video as they are, so the watermark stays with them.
Platform detection
Meta and TikTok also run their own checks on some uploads. Those labels come from the picture, not the file.
File names
A name like ChatGPT Image Oct 21.png tells people where the file came from. Rename it before you share it.
AI metadata remover FAQ
Data written into the file that records AI use: a C2PA manifest, an IPTC digital source type, AI credit lines, the AI app as creator tool and, in PNG files, the prompt. The AI label remover home page shows which apps write it.
Open this page in Safari, tap Choose photos or videos and pick from your library. The AI metadata is removed on your phone; tap Save to Photos to put the clean copy back in your camera roll.
Yes. MP4 and MOV videos lose their C2PA box, XMP AI fields and AI tags in the QuickTime metadata, without re-encoding.
Yes. Generative Fill and other Firefly features write a C2PA manifest with the source type compositeWithTrainedAlgorithmicMedia (edited with AI). It is removed; the camera details from your original photo stay.
Only if you ask. Turn on Also remove location and camera info, or use the metadata remover, which starts with it on.
Generative edits mark the whole file. Magic Eraser in Google Photos writes compositeWithTrainedAlgorithmicMedia even when it removes one small object, and Generative Fill in Photoshop does the same in C2PA. The AI metadata remover takes that mark out and keeps the rest of your photo’s data.
Yes. The AI metadata remover takes out the parameters, prompt and workflow text chunks that hold your prompt, seed and model, and copies the picture itself byte for byte.
No. Camera model, date taken, GPS, orientation and the color profile stay unless you turn on Also remove location and camera info, and so do camera capture credentials when nothing else has to change. An IPTC block that names AI goes whole, captions and copyright included.
Yes, for your own images. The EU AI Act asks AI providers to mark what their tools make and asks anyone who publishes realistic deepfakes to disclose them. Remove AI metadata from your own files and disclose fully AI-made content that looks real.
Yes. The AI metadata remover is free with no sign-up and no daily limit, and your files never leave your device.
Remove AI metadata from your photo
Drop in a photo or video, see every AI field and download a checked copy in seconds.