Rights and credits.

The rule

We harvest facts and rights-clear media. We never publish someone else's prose.

Everything else on this page follows from that one line. Museums, libraries and archives have spent two decades publishing structured, verified, rights-cleared records about millions of objects. A title, an artist, a date, a medium, a museum, an accession number: those are facts, and facts are free. The sentence that makes you care about the object is not a fact, and it is not free. So we take the first and we write the second.

This is not only caution. A card that reads like a museum label is not the product. Taking the scaffolding from the record means the whole of our writing effort goes into the part that is ours to do.

The gate

Four rules decide whether an object can be shown, in this order.

Per object, never per collection

An "open access" collection still contains works that are in copyright. We do not treat a collection-level policy as permission for anything inside it. Every institution we use publishes a per-object statement, and that statement is what we read.

Unknown rights are forbidden rights

An object whose record does not say, explicitly, that its image may be reused is treated exactly like one that says no. Silence is not permission, and a missing value never passes. This is enforced in the database rather than remembered by a person: a picture with no recorded permission cannot be rendered.

We rely on the institution's own words

Where a museum has released an image as CC0 or declared the work public domain, we act on that statement and store it as evidence, in the museum's own wording. We do not substitute our reading of case law for theirs, and we do not decide for ourselves that something old is probably fine.

The original record is kept

The provider's payload is stored unchanged alongside every object, so any question about rights that comes up later can be answered from the record we actually received, and not from our summary of it.

Where the works come from

Four institutions, as of September 2026. Each publishes an open access programme, and each marks the rights of an object on the object.

• The Metropolitan Museum of Art 752 works · Open Access, CC0
• The Cleveland Museum of Art 395 works · Open Access, CC0
• The Art Institute of Chicago 231 works · public domain, per object
• Rijksmuseum 180 works · public domain, per object

The gate is not decorative. The Rijksmuseum harvest alone skipped 20 objects the museum displays but does not hold the rights to, and 113 further records sit in a queue waiting for a person to look at why they were refused. Those objects are not in the product and never were.

We take nothing from sites without a published API or an open data programme. We do not scrape pages, and we do not take pictures out of search results.

Facts and identifiers

To know that the Rembrandt in one museum and the Rembrandt in another are the same man, you need an identifier rather than a name. Those come from Wikidata, whose structured data is released under CC0.

Wikipedia is a different thing and is treated differently. Its text is licensed in a way that would attach obligations to anything we wrote on top of it. It is also somebody's writing, which is the reason that would still hold if the licence changed tomorrow. We read it to check ourselves. We never copy from it.

Attribution

Where a licence asks for credit, the credit line and the licence are stored as required fields and shown next to the picture, not collected on a page like this one where nobody would look. A credits page is where attribution goes to be forgotten.

Every card also carries a citation back to the institution's own record for the object, so you can leave us and read the source. There are 3,285 of those citations in the product today.

Our own writing

The hook, the reason a work matters, the short story around it, the questions and their explanations: all of that is written for Arted. It is ours, and section 7 of the terms says what that means for you.

Some of it is drafted with machine assistance. Where that is true it is recorded as such against the item, in a field that exists for exactly that purpose, and it goes through the same fact checking, the same citations and the same publication gate as anything written by hand. Text assembled by a template is marked as a placeholder and is refused by that gate, so it cannot reach you dressed as editorial copy.

We would rather tell you this than have you find it out.

How we ask

The harvester makes one request per second per institution, identifies itself and a contact address in every request, backs off when it is asked to, resumes rather than restarts, and never re-fetches a record it already has. Several of these institutions permit far more than one request a second. We do not use it. A harvest that takes three nights is fine, and remaining welcome at these places is worth more than finishing early.

If you hold a right in something here

Gates fail. Records are wrong. An institution can mark an object as free to use and be mistaken about it, and we would have no way of knowing.

If you are the rights holder in a work, a photograph of a work, or a piece of writing you believe we have used without permission, write to us through the contact page and tell us:

• Which item, with a link or a title and an artist.
• What right you hold and how you hold it.
• How to reach you.

We will take the item out of circulation while we look, rather than after we have decided. If we were wrong we remove it, we say so, and we look at how the record got past the gate so that it does not happen to the next one. If we think we were right, we will tell you why and show you the institution's declaration we relied on.

The same route works for a correction. If a date, an attribution or a spelling is wrong, we would like to know, and fixing it costs nothing.

Credits

Typefaces

None are loaded. This page asks your device for the typefaces it already has, which is why it appeared as fast as it did and why no font company knows you were here.

Scripts

Two, both ours, both a few kilobytes. No framework, no analytics, no third-party code of any kind.

The mark and the illustrations

Drawn for Arted. The small illustrations on this site are the app's own, drawn as masks so that one file serves the light page and the dark one.

The store buttons

The glyphs are drawn here rather than borrowed. A store's badge is its trademark, and it is not ours to reproduce before we are in it.

The institutions

Listed in section 3. The product would not exist without their decision to publish openly, and that decision was neither cheap nor obvious.

The numbers on this page

Every figure here counts something in the product as it stood in September 2026, and every one of them can be produced from the database rather than estimated. When they change, this page changes and the date at the top moves.

The pool the product draws from is 1,558 works. Not every one of them has been written about yet: the front page says how many have.