Free image tools

Add AI training opt-out metadata to your images

Embed the IPTC Data Mining property in your JPEGs to state, in machine-readable form, whether they may be used for AI training. It is a rights signal that travels with the file, not a technical block.

This tool needs JavaScript. It runs entirely in your browser; nothing is uploaded.

Private by design. Your files are processed by your own browser on your own device. They are not uploaded, stored or seen by Archive.London. Tool terms

How to use the AI Training Opt-Out Metadata tool

  1. Choose your JPEGs or drop them on the page. Batches are supported.
  2. Select a Data Mining value, for example prohibited for AI/ML training or prohibited for generative AI/ML training.
  3. Optionally type an Other Constraints note describing your terms in plain words.
  4. Apply the metadata and download the files, or a ZIP for a batch.

To add an AI training opt-out to a photo, open the AI Training Opt-Out Metadata tool on Archive.London, drop in your JPEGs and choose a Data Mining value such as prohibited for AI/ML training. The tool writes the IPTC plus:DataMining property into each file's XMP without recompressing it, in your browser. It is a signal, not a block.

Choosing a Data Mining value

The values come from a controlled vocabulary published by the PLUS Coalition and adopted by IPTC. Choose the one that matches what you intend.

  • Prohibited for AI/ML training reserves the image against training of any machine-learning model.
  • Prohibited for generative AI/ML training is narrower: it targets generative models while leaving other machine-learning uses unaddressed.
  • Prohibited except for search engine indexing permits public search engines to find and index the image and prohibits other data mining.
  • Allowed states that data mining is permitted.
  • Unspecified records that no prohibition has been defined.

IPTC's own notes add a point that is easy to miss: the absence of a prohibition does not mean the owner grants permission. The optional Other Constraints text lets you describe further conditions in human-readable form.

A signal, not a lock: what this metadata can and cannot do

The Data Mining property is a declaration. It does not encrypt, watermark or alter the picture, and it cannot physically stop a crawler from downloading the file. It works only when the party collecting images reads embedded metadata and chooses, or is obliged, to respect it.

In practice many pipelines do neither. Social platforms, content delivery networks and image optimisers frequently strip XMP, and a scraper may discard metadata before anyone looks at it. A copy of your picture re-saved by someone else will usually have lost the field.

It is therefore best treated as one layer among several: clear terms on your website, crawler directives such as robots.txt, a Web Statement of Rights in the file, and a record of what you published and when. Each catches cases the others miss.

The standard and the law behind the field

The property was added in IPTC Photo Metadata Standard 2023.1, approved by IPTC members on 4 October 2023, and is stored in XMP as plus:DataMining with a value from the PLUS vocabulary.

Its legal relevance is clearest in the European Union. Article 4(3) of Directive (EU) 2019/790 of 17 April 2019 on copyright in the Digital Single Market allows text and data mining of lawfully accessible works unless the rightholder has expressly reserved that use in an appropriate manner, such as machine-readable means for content published online. Embedded metadata is one candidate for such a reservation, though whether a particular method is sufficient is a matter for courts and regulators.

Elsewhere the position differs, and IPTC itself notes that regional laws may overrule the chosen value. Legal effect varies by jurisdiction and is not guaranteed. This is not legal advice.

Rights reservation and licensed training data are two sides of one question

Reserving your rights does not have to mean refusing every AI use. It establishes that use requires your agreement, which is the starting point for any licence. Developers who want training data with a clear chain of title increasingly look for collections where one party can grant permission for the whole body of work.

The Robert Harper Photograph Library is an example of that structure: 95,487 photographs by one photographer, held by a single rights holder. Licensing for machine-learning use is available by negotiation, subject to rights review, and the approach is set out on the AI training data licensing page.

Frequently asked questions

Does this stop AI companies from using my images?

Not by itself. It embeds a machine-readable statement that you reserve your rights against data mining or AI training. Whether a developer honours it depends on their systems reading the field and on the law that applies to them. Some pipelines strip or ignore metadata entirely. Treat it as a clear, standardised signal of your position that supports other measures.

Is the IPTC Data Mining opt-out legally binding?

That depends on where you and the user of the image are. In the EU, Directive 2019/790 lets rightholders reserve text and data mining by appropriate machine-readable means, and embedded metadata may serve as such a reservation. Other countries take different approaches. The legal effect varies by jurisdiction and is not guaranteed, so take professional advice if it matters to you. This is not legal advice.

Will the opt-out survive when I upload the image to social media?

Often not. Many social networks and image hosts re-encode uploads and remove XMP metadata, so the copy they serve may no longer carry the field. It is most likely to survive on your own website, in direct file transfers and in agency or library delivery systems that preserve metadata. Test a platform by uploading, downloading the result and checking it in a metadata viewer.

Does adding the field change or recompress my photo?

No. The property is written into the JPEG's XMP metadata segment and the compressed image data is left exactly as it was. Quality, dimensions and colour profile do not change and the file grows only slightly. The tool works on JPEG files. Processing takes place in your browser, so your images are not uploaded anywhere.

How can I check whether an image already has a Data Mining value?

Open the file in the EXIF and Metadata Viewer. If the XMP contains plus:DataMining, the viewer highlights it with the other rights fields and shows the value that was set. If nothing appears, either the field was never added or the metadata has been stripped somewhere along the way. Absence should not be read as permission.

Licensing the archive for AI and data projects

95,487 photographs from one photographer and one rights holder. Read the dataset card, search the 503-folder inventory, discuss a sample set under NDA, or arrange a first look as scanning proceeds.

See AI and data licensingWrite to the archive