AI Training Opt-Out in Image Metadata: Mark Your Photos "Not for AI"
IPTC, the body behind the photo metadata standard, defines a Data Mining property for exactly this: a field in the file that says whether it may be used to train AI. Setting it to "Prohibited for AI/ML training" puts your terms inside every copy of the image, next to your name and copyright notice.
This page opens the Metadata Remover with Add your details switched on and the AI-training opt-out ticked. Location and camera data are removed, then the flag, your credit and your licence are written in, and each file is read back to confirm the flag is there. Be clear about what it does: it is a statement of your terms that honest crawlers and licensing systems can read, not a lock. It cannot stop anyone who ignores it, and sites that strip metadata remove it along with everything else.
Metadata Remover: Remove EXIF, GPS & Hidden Data from Your Files
Find and permanently remove hidden metadata from images, documents, video, audio and vector files: EXIF, GPS, XMP, IPTC, MakerNotes, ID3 tags, Office properties and QuickTime atoms. Deep Clean goes further in Word, Excel, PowerPoint, OpenDocument and PDF files and removes tracked changes, comments, speaker notes, hidden sheets, attachments and scripts. Every byte is parsed and rewritten in your browser, and nothing is uploaded. We never fetch map tiles either, because that would leak the very coordinate you are erasing.
How it removes metadata (per format)
JPEG / PNG / WebP: the EXIF block is rewritten with only the tags you allow (drop GPS, keep orientation); XMP, IPTC and comments are removed as chosen. Pixels are untouched.
TIFF / HEIC / AVIF: these can’t be edited byte-safely, so the pixels are re-encoded to a clean PNG/JPEG. All metadata goes and the format changes.
PDF: the Document Info dictionary (custom keys included), XMP on the document, pages and images, and the file ID are removed. Deep Clean also drops JavaScript and launch or submit actions, attachments, comments and markup, XFA form data and editor data (PieceInfo). Objects nothing points to any more are pruned, so removed data is out of the file, not just unlinked, and incremental-update history collapses into one revision.
DOCX / XLSX / PPTX:docProps/core.xml, app.xml and custom properties are cleared. Deep Clean accepts tracked changes and removes comments, hidden text, speaker notes, hidden sheets, custom XML and the saved preview, and local file paths in links are cut to the file name.
ODT / ODS / ODP:meta.xml authors, dates, editing time and custom fields, the preview thumbnail and the printer name go; Deep Clean also accepts tracked changes and removes comments and speaker notes.
MP4 / MOV / M4A:udta/meta boxes (including QuickTime GPS) are zeroed in place, so playback is unaffected.
MKV / WebM: title, dates, writing app and Tags are neutralised in place as EBML Void elements.
MP3 / WAV / FLAC: ID3v2, RIFF INFO and Vorbis comments are rebuilt without the removed fields; ID3v1 and cover art are dropped.
SVG / EPS:<script> elements, on* handlers, editor namespaces, RDF metadata and author/date comments are stripped.
Add your details after cleaning
Switch on Add your details to sign each cleaned file with your name, copyright notice, credit line, title and licence. It runs as a second pass over the cleaned bytes, so nothing that was removed can come back. JPEG files get the details in XMP, IPTC and EXIF, which covers Google Images (including the Licensable badge when you add a licence URL), Adobe apps, Windows and macOS; PNG and WebP get XMP, EXIF and, for PNG, its own text fields; PDF gets Author, Title and XMP; Word, Excel, PowerPoint and OpenDocument files get Author, Title and custom properties. Text in any language is stored as UTF-8, and EXIF, which only holds plain Latin text, is used only when the text fits. Fields accept {year}, {filename} and {n}, profiles are saved in your browser, and every file is read back to list the readers that will show the credit.
The optional AI-training opt-out writes the IPTC Data Mining flag “Prohibited for AI/ML training”. It states your terms; it does not stop anyone technically. Instagram, Facebook and X strip metadata on upload, so the details stay with the files you send and publish yourself. GIF, SVG, EPS, audio and video files are cleaned but not signed.
Limits
This tool edits metadata and hidden content. It does not run OCR and cannot remove what is part of the visible content, such as text under a black box or faces in a photo. PDF form answers are kept, and PDF layers that are switched off are flagged for you to check rather than deleted. A formula that reads from a hidden Excel sheet shows #REF! once that sheet is removed. MP4/MOV and Matroska tags are cleared as a block, not field by field. Browsers stamp a downloaded file with the time it is saved, so “Preserve file dates” only controls the timestamps embedded in the file.
Frequently asked questions
What exactly is written into the file?
The XMP property plus:DataMining with the value http://ns.useplus.org/ldf/vocab/DMI-PROHIBITED-AIMLTRAINING, as defined by IPTC and PLUS. It says the file may be indexed by search engines but not used to train AI or machine-learning models. Your creator, copyright and licence fields go in alongside it if you fill them in.
Will this stop AI companies from training on my images?
Not by itself. It is a machine-readable statement of your terms, in the place the photo industry agreed to put it. Companies that respect opt-outs can find it, and it records your intent on every copy, which can matter in a licensing or legal dispute. It does not technically prevent copying, and it is gone from any copy a platform has stripped.
Should I also block AI crawlers on my website?
Yes, if you publish your own work. The file flag travels with each image; a robots.txt rule for AI crawlers and the site's own terms cover the pages. The two work together, and neither replaces the other.
Which file types can carry the opt-out?
JPEG, PNG, WebP, HEIC, AVIF and TIFF (the last three are converted to JPEG or PNG) and PDF carry it in XMP. Word, Excel, PowerPoint and OpenDocument files get it as a custom document property named "AI training". GIF, SVG, audio and video files are cleaned only.
Is my photo uploaded anywhere to be checked?
No. The whole check runs in your browser and the file is never uploaded. Previewing and checking need no account; an account is needed only to spend tokens when you download the result.
Key facts about Metadata Remover: where it runs, what it costs, what it accepts and produces, and its limits.
Where it runs
Entirely in your browser. Files are never uploaded.
Cost
20 tokens per file (about $0.20 at the base rate); 250 tokens for up to 50 files
Accepts
JPEG, PNG, WebP, GIF, TIFF, HEIC, AVIF, PDF, Word/Excel/PowerPoint, ODT/ODS/ODP, MP4, MKV, MP3, WAV, FLAC, SVG, EPS, a folder or a ZIP
Produces
The same file with its metadata removed and, with Deep Clean, its tracked changes, comments, speaker notes, hidden sheets, PDF attachments and scripts too, plus a before/after field report. Batches download as a ZIP that keeps your folders, with optional neutral file names.
Batch
Up to 50 files per run.
Account needed
Only to spend tokens. Browsing and previewing are free.
What it does not do
It removes metadata and hidden content, not visible content. Text, faces or a location shown on the page or in the image itself stay there, and PDF form answers are kept.
Metadata Remover: before and after
An illustration with sample files; your own files show their own results.
