← All guides For creators

Stock photo sites and AI submissions

Libraries went from banning generated work to selling it. What each side of the marketplace needs to check now, and why undisclosed submissions are the real problem.

· 10 min read · Best AI Image Detector

Every major library now has a position, and they differ. The consistent rule across all of them is disclosure: an undisclosed generated submission is what gets a contributor account closed, not the use of generation itself.

How the market changed position

The first reaction across the industry was prohibition. Libraries banned generated submissions outright, citing unresolved questions about training data and about who owned the result.

That position lasted about a year. Buyers wanted the imagery, contributors were producing it anyway, and several libraries launched their own generators trained on content they had already licensed, which made a blanket ban untenable.

What replaced it is a set of differing policies with one common feature: generated content is acceptable under conditions, and the conditions always include declaring it. The differences are in whether editorial use is permitted, whether model releases are required, and how the work is labelled to buyers.

The result is a marketplace where the same image can be compliant on one platform and grounds for removal on another, which is why contributors get caught more often through inattention than through intent.

What a contributor has to get right

Three things, and only the first is about the image itself.

  1. Declare it. Every library that permits generated content requires it to be marked as such at submission. This is where accounts are lost.
  2. Do not submit generated likenesses as editorial. Editorial imagery asserts that something happened, and a generated scene asserting that is the most serious version of this.
  3. Watch the release requirements. A recognisable face, even one that does not exist, raises questions some libraries handle by requiring documentation nobody can provide.

The second point causes the most damage when it goes wrong. Editorial licensing exists specifically for images used to illustrate real events, and a generated image entering that pipeline can end up in a news publication as an apparent record of something.

The third is more subtle than it looks. A generated portrait has no model to release, which some libraries treat as a reason to reject it and others treat as a reason to accept it. Read the specific policy rather than assuming.

What gets submissions removed
  • Undeclared generated content The single most common cause of account action.
  • Generated imagery in an editorial category Asserts that something happened. Treated seriously everywhere.
  • Prompted in the style of a named artist Raises a claim the library does not want to defend.
  • Recognisable trademarks or logos An old rule that generation makes easy to break accidentally.
  • Near-duplicate flooding Twenty variations of one prompt is a volume problem, not a portfolio.
Where a contributor account is actually at risk. Only the first row is about generation.

What a buyer should check

The label on a stock page tells you what the contributor declared, which is not always what they did. For most uses that gap is irrelevant; for a few it matters a great deal.

When the distinction is worth verifying
UseDoes it matter?Why
Blog header or backgroundNoDecorative, nobody is misled
Advertising a physical productYesThe image implies the product looks like this
Illustrating a news storyYesAsserts that something happened
Case study or testimonialYesImplies a real customer
Concept art or abstractNoNobody expects a photograph
Medical or scientific referenceYesA plausible invention teaches something untrue

The pattern across the yes rows is consistent: verification matters wherever the image is doing the work of evidence rather than the work of decoration. That is the same line that runs through every article on this site.

The opportunity for photographers

It is worth ending on the commercial reality rather than the compliance one. Generated imagery has made generic pictures nearly worthless and specific ones more valuable, because the things a prompt cannot produce are exactly the things buyers now struggle to source.

A photograph of a real place on a real day, of a named person who consented, of a specific piece of equipment in a working environment: none of these can be prompted, and all of them are what a buyer needs when the image has to be evidence.

Provenance is part of that value. A photographer who can supply a raw file with intact metadata, and increasingly a signed credential, is selling something a generator structurally cannot offer, and libraries are beginning to price that.

The practical advice for contributors is therefore not to avoid generation but to be clear about which business they are in. Decorative imagery competes on volume and price; evidential imagery competes on provenance, and only one of those has a floor.

What contributors should do about older work

Portfolios built over several years often contain images the contributor no longer remembers the production history of, particularly where generative fill entered mainstream editing software quietly and got used for ordinary cleanup.

A retroactive audit is uncomfortable and better than the alternative. Libraries are running detection over existing catalogues, and a contributor who identifies and re-declares their own borderline work is in a very different position from one who waits to be told.

The practical approach is to sort by risk rather than by date. Editorial submissions first, then anything with a recognisable person, then commercial imagery of identifiable products. Decorative and abstract work can wait, because nobody is misled by it.

Where a file cannot be accounted for, the honest move is to withdraw it rather than to re-declare it as something you are guessing about. A smaller portfolio you can stand behind is worth more than a larger one you cannot.

One structural point is worth adding. Libraries are commercially motivated to police this properly, because a buyer who licenses a generated image believing it to be a photograph has a complaint against the library rather than against the contributor. That alignment is why enforcement has tightened quickly.

It also explains why disclosure rather than prohibition became the settled answer. A library that knows what it is selling can price it, label it and licence it appropriately, and a contributor who tells them is solving the library problem rather than creating one.

Contributors sometimes ask whether declaring generated work hurts sales. The evidence so far suggests it segments them rather than reducing them: buyers looking for decoration do not filter it out, and buyers who need a photograph were never going to be satisfied by an undeclared generation anyway.

Questions people ask

Do stock sites accept AI images now?
Most major libraries do, under conditions that differ between them. The universal requirement is disclosure at submission. Some restrict generated work from editorial categories entirely, and some have additional rules about recognisable people, so read the specific policy rather than assuming.
What gets a contributor account closed?
Undeclared generated submissions, by a wide margin. Libraries run their own detection over portfolios, and a contributor with a mixed portfolio who declared none of it looks worse than one who declared all of it. The disclosure is what the rules are about.
Can I sell generated images as editorial?
Generally no. Editorial licensing exists for images illustrating real events, and a generated scene entering that pipeline can end up in a news publication as an apparent record of something that did not happen. This is treated as the most serious version of the problem.
As a buyer, should I check what I license?
It depends on the use. For a decorative header, no. For advertising a physical product, illustrating a news story, or anything implying a real customer or a real event, yes, because the image is doing the work of evidence and the label reflects what a contributor declared.
Is photography still worth doing commercially?
Generic photography has been badly devalued and specific photography has become more valuable. A real place on a real day, a named person who consented, a particular machine in a working environment: none of that can be prompted, and it is what buyers need when an image must be evidence.
Should I keep raw files?
Yes, and increasingly it is a commercial asset rather than an archive habit. A raw file with intact metadata is the strongest available answer to any question about a photograph, and libraries are starting to treat verifiable provenance as something worth paying for.