AI stock images in your online store: do you have to label bought or licensed AI images?

Research as of 1 October 2026. This guide covers AI images you do not create yourself: AI-generated images from stock libraries such as Adobe Stock, Shutterstock, Getty Images, iStock, Freepik or Depositphotos, and images that an agency, photographer or supplier creates with AI and delivers to you. The basics of the labelling obligation, meaning your role, the image types and the form of the label, are in the guide Do I have to label AI images in my online store?. The transparency obligations under Article 50 are explained in a separate guide.

Who has to label an AI stock image?

Article 50(4) of the AI Act (Regulation (EU) 2024/1689) does not oblige everyone who shows an AI image. It addresses “Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake”. What counts is who used the AI system, not who publishes the finished image. The typical cases:

In each of these cases: if an image shows your product differently from what you deliver, no label will help. More on that below.

What does the AI Act itself say?

Article 3(4) defines a deployer as “a natural or legal person, public authority, agency or other body using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity”. Article 50(4) ties the disclosure to exactly this use, and recital 134 refers to “deployers who use an AI system to generate or manipulate image, audio or video content”.

If you license a finished image, you use the image, not the AI system that generated it. On its wording, the definition of a deployer therefore does not fit a mere licensee. The Regulation does not, however, contain an express rule on stock images.

Article 25, which governs when a distributor, deployer or other third party is treated as a provider, only concerns high-risk AI systems (paragraph 1) and plays no role for images in a store. According to the guidelines, a company only becomes the provider of an image generator if it modifies an existing system, for example with new training data, and puts it into service under its own name (para. 11). Buying a stock image has nothing to do with that.

The Digital Omnibus (Regulation (EU) 2026/1744) changes none of this: Article 3 is amended only in point 14 and supplemented by points 14a and 14b (Article 1 point 4), Article 50 only in paragraph 7 (Article 1 point 20), and the new versions of Article 25(2) and (4) (Article 1 point 12) still only concern high-risk systems. A consolidated version of the AI Act was not available on 1 October 2026.

What do the Commission guidelines say?

The European Commission’s guidelines on Article 50 (annex to C(2026) 5054 final of 20 July 2026) do not mention stock images, but they draw a clear line.

• Para. 12, what “under its authority” means: assuming responsibility “over the decision to deploy the system and over the manner of the actual use of the system (including its outputs)”. Technical control over the system is not required.
• Para. 16, who is not a deployer: actors whose role is limited to disseminating or transmitting AI content created by third parties, “or who receive or are exposed to AI-generated or manipulated content without having authority over the use of the AI system for the purpose of the AI content generation or manipulation, are not deployers within the meaning of the AI Act”. The examples given in para. 16 are hosting services, online platforms and broadcasters. They are “strongly encouraged to preserve the marking and labelling”.
• Para. 17, using content does not make you a deployer: actors who disseminate content in their professional capacity are encouraged to take proportionate measures so that people can recognise its artificial origin. And: “However, the mere fact that an actor labels content or uses labelled content does not mean that the latter qualifies as a ‘deployer’ within the meaning of the AI Act.” Para. 17 also notes that such actors may be required to label under other legal or deontological rules.

If you license an AI stock image, you did not decide on the use of the generator. Under paras. 16 and 17 you are therefore not a deployer of that system, and on the most natural reading this remains so even if you know that the image was made by AI or searched specifically for AI images. Applying this to stock images is still interpretation: the examples in para. 16 are platforms and broadcasters, not retailers, and under para. 5 the guidelines are non-binding; only the Court of Justice of the EU can interpret the AI Act authoritatively.

What does the AI Service Desk at RTR say?

Of the bodies checked, only the AI Service Desk (KI-Servicestelle) at RTR, the Austrian regulator for broadcasting and telecommunications, addresses stock images expressly. Asked “I obtain unedited and published photos from stock photo platforms. Do I have to add a label to these photos?”, its FAQ (accessed 1 October 2026) answers, translated from German:

• “No, in this constellation you have no labelling obligation of your own under the European Commission guidelines: anyone who merely disseminates or uses AI content generated by third parties, without having authority over the generating AI system themselves, is not a ‘deployer’ within the meaning of the AI Act and is therefore not subject to the disclosure obligation for deep fakes (para. 16 of the Article 50 guidelines).”
• “You are, however, encouraged to preserve existing labels and machine-readable markings (e.g. metadata or watermarks) and not to remove them (para. 16 of the Article 50 guidelines).”
• “The situation is different if you edit or alter the photos you obtained with an AI system yourself.” Images that are deep fakes then have to be disclosed. “Irrespective of the AI Act, a voluntary label can make sense to avoid misleading people.”

The AI Service Desk informs and advises; it is not a supervisory authority (more in the guide Labelling AI images in Austria). Its answer relies on the guidelines, which apply across the EU, but it remains the view of an Austrian information service, not a legal rule and not a decision.

On 1 October 2026, there was no statement on bought or licensed AI images in the Commission’s FAQ on Article 50 (page as of 24 July 2026), on the Bundesnetzagentur’s pages on the transparency obligations, or in the guide of the Wettbewerbszentrale, the German industry body against unfair competition (version 2.0, as of 29 July 2026). The Wettbewerbszentrale describes the deployer there as whoever uses an AI system under their own responsibility; the definition is broad and covers everyone who uses an AI system in their work. It, too, ties the role to use of the system.

When are you a deployer after all?

You edit the image with AI yourself. Under the guidelines, manipulation refers to existing content, “whether synthetic or not”, that an AI system alters in accordance with human instructions (para. 59). If you use generative fill to put in a different person, insert your product into a stock scene with AI, or let AI extend the scene, you are the deployer for that edit. Whether the result needs a label depends on the deep fake test, as with any other image (Which images in your store need an AI label?). Under para. 116, colour correction, noise reduction, upscaling or a background replacement for clearly aesthetic purposes usually have only a minor impact and do not normally turn an image into a deep fake.

You create the image with the library’s generator. Many stock libraries offer their own AI generators alongside the library. An image you create there from a prompt is not a stock image: you are using the AI system yourself and you are the deployer. Freepik says so expressly in its terms of use (June 2026 version, section 14.2): “For the purposes of the AI Act, Users of the AI Products act as deployers of AI systems.” According to its licence agreement, Depositphotos does not add images from its generator to the shared library (section 6, effective 12 September 2025); they are your own outputs as well.

You steer an agency. See the next section.

What applies to agencies, photographers and suppliers?

Para. 14 of the guidelines deals with this case expressly, in two directions:

• “A legal person remains a deployer even if it involves third parties (e.g. contractors, freelancers) in the operation of the system on its behalf and under its responsibility and control.”
• “By contrast, a company that merely commissions an advertising agency to produce an advertisement, without taking decisions and exercising control over whether and how the advertising agency uses AI in the production process, is not a deployer.”

So if you specify that the images are to be made with AI, or have a say in how, for example through prompts, reference images or choosing between AI variants, there is a strong case that you remain the deployer (our reading; para. 14 does not draw the line in detail). If, on the other hand, the agency delivers finished images whose creation you had no say in, it uses the AI system under its own authority; on the most natural reading, the agency is then the deployer. The guidelines expect deployers in production and distribution chains to take proportionate measures so that their label reaches the audience, for example through contractual conditions with distribution partners (para. 12). That is how a labelling obligation can still end up with you, under the contract rather than under Article 50.

The agency’s side, with all contract building blocks, is covered in Creating AI images for store clients. In practice, every order to an agency, photographer or retoucher should therefore cover:

1. whether AI may be used, and for what,
2. a list of which delivered images were created wholly or partly with AI, and with which tool,
3. who labels, and with which variant (AI GENERATED or AI MODIFIED),
4. delivery of the original files with unaltered metadata,
5. an assurance that the images show your product as you deliver it.

For manufacturer’s and supplier’s images that you use unchanged, the same applies under para. 16 as for stock images: you did not decide on the use of AI (our own reading). You still advertise with the image yourself and are responsible for a misleading product presentation.

What applies regardless of the label?

Article 50(6) of the AI Act leaves other transparency obligations unaffected. The most important limit for stock images is therefore drawn not by the AI Act but by unfair competition law:

• Product. An AI stock image of a similar product used as a product photo does not show the goods you deliver. In Germany, Section 5(2) no. 1 UWG treats a commercial practice as misleading if it contains untrue or deceptive statements about the essential characteristics of the goods, such as their type, design or quality. In Austria, Section 2 UWG applies; in Switzerland, Article 3(1)(b) UWG. An AI label does not cure such misleading content; the image needs replacing.
• Business. A realistic AI stock image presented as “our team”, “our workshop” or as a customer photo next to reviews makes a statement about your business that is not true. In Germany, Section 5(2) no. 3 UWG expressly covers statements about the person, characteristics or rights of the trader. Whether such an image is misleading in a given case has not been decided (our own reading); a label alone does not turn an invented team into a real one.
• Context. Under the guidelines, whether an image appears authentic depends on the context and the foreseeable audience, without any intention to deceive being required (para. 114). The same stock image can be harmless on a topic page and problematic next to a product.

When should you label a stock image anyway?

The fact that, on this interpretation, Article 50(4) does not oblige you to label an unchanged stock image does not make a label wrong. There are four reasons for a voluntary label:

1. The interpretation is not settled. The guidelines are non-binding (para. 5), RTR is not a supervisory authority, and there is no court ruling.
2. The guidelines also encourage non-deployers. Actors who disseminate content professionally should help people recognise its artificial origin (para. 17); RTR considers a voluntary label sensible “to avoid misleading people”.
3. The bodies that send warning letters read the rules broadly. The Wettbewerbszentrale recommends in its guide that AI images should be labelled as such when in doubt, and considers infringements of the AI Act a possible infringement of the German UWG as well; the Austrian Schutzverband gegen unlauteren Wettbewerb follows this guide.
4. A label costs little. You do not have to replace the image for it.

This leads to a rule of thumb. It is this guide’s reading, not the legal position:

If you label a fully generated stock image, the “AI GENERATED” icon fits; variants and placement are covered in The EU icons for AI images.

How do you recognise AI stock images?

1. From the library’s marking. The providers take different approaches. According to their own statements, accessed 1 October 2026:

• Adobe Stock accepts AI images from contributors. When uploading, they must tick “Created using generative AI tools”, which is “Required for all content created with generative AI software” (contributor guidelines, as of 11 June 2026). The product terms refer to stock assets “labeled as ‘Generated with AI’” (version of 16 January 2026). The marking therefore relies on what the contributor declares.
• Shutterstock does not accept AI images from contributors under its own policy (as of 16 July 2025), but puts images from its own generator into the library: they are “added to Shutterstock’s vast collection of content to be licensed by other users” (contributor FAQ, as of 18 April 2024). Its search has an “AI Generated” filter that excludes AI images or shows only them (Help Center, archived 2025 version; the current page blocks automated access).
• Getty Images tells contributors: “We don’t accept content created by generative AI” (requirements of May 2025). iStock does not accept imagery created with generative AI into its library; images from its own AI tools are, by its own account, clearly marked as AI-generated and not added to the library.
• Freepik requires contributors to add a keyword to AI content; according to its contributor blog, the search can filter AI results by it (post of 20 March 2024). How the marking appears to buyers could not be verified.
• Depositphotos does not accept AI-generated files from contributors, according to its FAQ; its search nevertheless has an “AI Content” filter with an “Exclude Generative AI” option.

No obligation to label licensed AI stock images for your customers was found in the accessible terms of these providers. Some, however, prohibit deceptive use, for instance Shutterstock (use “in a … deceptive context”) and Depositphotos (use that would lead to a complaint about deceptive advertising). Filters and markings depend on what contributors and providers declare, so an image without an AI note is not necessarily a real photo. Canva’s library information could not be retrieved on 1 October 2026.

2. From the marker in the file. Many generators write a provenance marker (C2PA, IPTC DigitalSourceType) into the file. Whether a stock library keeps it in the licensed file has not been measured in the Easy AI Act Image Labels test bench. Whether your file carries one is shown by the free AI image checker; the file does not leave your device. A marker found is evidence of AI origin; a missing marker says nothing either way. More on this: Content Credentials and C2PA.

3. From the licence. Your invoice or licence history names the image ID. It leads you to the detail page and therefore to the library’s marking, even years later.

4. From the sender. For agency, photographer and supplier images, the only way is to ask, ideally with the contract clause above.

An app such as Easy AI Act Image Labels reads such markers on the original in the Shopify media library and suggests a label for matches; whether a stock image without a marker gets a label is your decision.

How do you document stock images?

As a mere licensee, the expectation in para. 148 of the guidelines that you can demonstrate compliance does not apply to you, because it is addressed to deployers. A short list is still worth keeping: it shows why you assessed an image as you did, and it helps if a library changes its marking or a competitor asks. For each image:

• stock library, image ID and link to the detail page,
• licence: type, date, invoice number,
• the library’s AI marking, ideally as a dated screenshot,
• marker in the file yes or no, with the date checked,
• your own editing: none, or tool and type of edit,
• where the image appears in your store,
• decision on the label, yes or no, with a short reason.

If you have edited a stock image with AI yourself, you are the deployer for that edit; para. 148 then applies as for any other AI image (Who enforces the EU AI image labelling rule in your store?).

What about older stock images?

Older images are treated like new stock images at first: used unchanged, you are not a deployer on the interpretation above, and the cut-off date plays no role for you. It matters if you have edited an older image with AI yourself. The guidelines say in para. 154 that deep fakes generated or manipulated before 2 August 2026 do not need to be labelled retroactively; at the same time they encourage “deployers of AI systems and other actors who are in the possession of or disseminate pre-existing unlabelled deep fakes” to do so, without disproportionate efforts such as auditing entire content databases. RTR reads this more strictly: according to its FAQ, the transparency obligations apply if content generated before the cut-off date is published or first used on or after 2 August 2026. Details: Do AI images from before 2 August 2026 need a label?

In practice: check older licences via the image ID. If the library now shows an image as AI-generated, classify it using the table above. If you do not know where an old image came from, the motif decides: label a realistic image close to a product when in doubt, not a clearly artificial one.

What is still open?

• Binding force: the guidelines do not mention stock images expressly and are non-binding (para. 5). RTR’s answer is the view of an information service. No statement on stock images could be found from the Bundesnetzagentur, the market surveillance authority in Germany, on 1 October 2026. No court ruling or published case on labelling AI images in online retail was known as of 30 September 2026 (tracker October 2026).
• Who labels towards your visitors: whoever creates an AI image professionally and sells it as a stock image is likely a deployer themselves (Article 3(4); our own reading). Whether their disclosure has to reach the visitors to your store is open. The guidelines point deployers in distribution chains to contractual conditions (para. 12); whether and how stock libraries build this into their licences is worth checking in each licence.
• Outputs as part of authority: para. 12 includes the system’s outputs in the deployer’s “authority”. Anyone who wanted to infer that using a third party’s AI image under your own responsibility is already enough to make you a deployer would have to set aside paras. 16 and 17; only a binding decision would rule out that reading.
• Where the line lies for agencies: the guidelines do not say when a brief, reference images or choosing between AI variants amount to deciding “how” AI is used under para. 14.
• Unfair competition law: no court has decided whether a missing label on a realistic stock image is unfair in itself.

In short

An AI stock image that you only license and show unchanged does not make you a deployer under the Commission guidelines and RTR’s answer; Article 50(4) then does not oblige you to label it. You become a deployer if you alter the image with AI yourself, use the library’s generator yourself or tell an agency whether and how to use AI. No label helps an image that misleads about your product or your business. For realistic motifs close to a product, there are good reasons for a voluntary label; if you record the source, licence and decision for each image, you can explain your assessment later.