Call · 30 min
MarketingMattia EspositoUpdated 11 November 202612 min read

AI images in advertising. Yours by contract, protected only if there is an author, guaranteed almost never: what the law and the generators' terms say.

Paying for the generator gets you the image, but it does not protect it in law or cover you if it infringes someone else's rights. Here is what the law and the terms say, and the ad checklist in Word and PDF.

In brief

A business can use AI-generated images in its ads: the terms of the most widely used generators allow it and no Italian law forbids it. The limits lie elsewhere: faces of real people, other companies' trademarks, realistic images needing a label.

Owning the image does not mean it is protected. Since 10 October 2025, Italian copyright law, as amended by Law 132/2025, protects works of human creativity, including those made with AI, if they are the result of the author's intellectual work. How much work is enough, the law does not say.

Almost nobody guarantees you. Among the terms read on 10 November 2026, the only written indemnity is Adobe Firefly's, on business plans and up to 10,000 dollars per output.

The eight-line checklist is free to download below, in Word and PDF, with no email required.

Written by: Mattia Esposito, AI and marketing consultant, founder of Itria AI. From Puglia, based in Turin since 2023, he works remotely in Italy and abroad. LinkedIn profile.

The checklist for an ad with an AI image, free to download

Eight checks to run before publishing an ad with a generated image, with where to look and space for the result, plus the table of generator terms and a log. Word to fill in, PDF to print. Every line points to a public text; in doubtful cases the answer belongs to a lawyer.

The eight-line checklist for an ad with an AI-generated image, with the table of generator terms
FileContentsLink
Wordto fill in, in Italian

The eight checks, the generator table, the image log.

immagini-ai-pubblicita-diritto-autore-checklist.docx

PDFto print, in Italian

The same checklist, to tick by hand before approving the ad.

immagini-ai-pubblicita-diritto-autore-checklist.pdf

N.What to checkWhere it comes from
1

The generator and plan allow commercial use and leave you ownership of the output.

Generator terms

2

The input contains no photos, logos or works belonging to others.

Canva and Adobe terms

3

No recognisable face without written consent.

Art. 96, Law 633/1941

4

No trademark or product of another company in view.

IAP Code, art. 13

5

The product is shown as it really is.

Rules on misleading advertising

6

If the image looks real, a label and the platform's disclosure setting.

AI Act, art. 50

7

The human contribution is documented and saved.

Art. 1, Law 633/1941

8

Indemnity checked, log filled in.

Plan terms

For more than two years I posted other people's paintings on @artartism, the Instagram page I started at fourteen, with the artist's name and credits under every post. When I went back through the data, I put it like this: “You've already seen the famous one. And you've already scrolled past it.” Klimt's The Kiss got 1,900 likes, a living fantasy painter 15,300. The full post.

I think of it every time I see an ad with a famous face generated by AI. Legally, as you will read below, nothing is riskier, and on my page the famous name did not even perform best.

In August, preparing Itria's first Meta campaign, I read about a hundred ads in a week. Images take a minute to generate. What decides whether an ad holds up comes earlier: the idea, and then the checks on this page. Competitors' ads can be read in the Meta Ad Library, and the opening that makes people stop is covered in the guide to hooks and headlines.

Can I use AI-generated images in advertising?

Generally yes: no Italian law forbids using an AI-generated image in an ad, and the terms of the most common generators allow it. The real questions are three, and they often get mixed up: who owns the image, whether the law protects it, and who pays if it infringes someone else's rights.

Three different questions about the same generated image: who owns it by contract, whether the law protects it, who pays if it infringes a right The generated imagefor your ad Whose it isthe generator's contractusually: yours Is it protectedthe law: art. 1, Law 633only with human input Who pays if it infringesthe indemnity, if anyusually: you
Three questions that often get confused. Each is answered by a different source: the contract, the law, the indemnity clause.

Each one has its own source. The generator's contract says who owns the image. Copyright law says whether it is protected. The indemnity clause, where there is one, says who pays. A fourth rule sits alongside: if the image is realistic enough to look genuine, from 2 August 2026 article 50 of the AI Act may require you to disclose it, as explained on the page on the AI Act and generated content.

Can AI images be copyrighted? Italy's Law 132/2025

In Italy an image made with AI is protected only if it is the result of a human author's intellectual work. This is article 1 of Law 633/1941 on copyright, as amended by article 25 of Law no. 132 of 23 September 2025, in force since 10 October 2025.

The current text protects “opere dell’ingegno umano” (works of human creativity) of a creative character, “anche laddove create con l’ausilio di strumenti di intelligenza artificiale, purché costituenti risultato del lavoro intellettuale dell’autore”, meaning even when made with the help of AI tools, provided they result from the author's intellectual work. How much human contribution is needed, the provision does not say: the courts will decide case by case.

Two practical points follow. An image produced from a prompt and published as it came out might have no copyright protection, and a competitor could reuse it. Other protections still apply, such as trademarks and unfair competition, which have nothing to do with copyright. If you may one day want to defend the image as yours, keep sketches, choices between variants, retouching and working files: line 7 of the checklist.

Law 132/2025 also adds article 70-septies, which allows text and data mining with AI systems within the limits of articles 70-ter and 70-quater, respecting the rights holders' reservation. It concerns those training the models, not the business using the image in an ad. For breaches, article 26 of the law introduces a criminal penalty, letter a-ter) of article 171.

ChatGPT, Gemini, Canva, Firefly and Midjourney: can I use AI images commercially?

Almost every generator leaves the output to its creator, but none guarantees that it is unique or that it does not infringe anyone's rights. OpenAI's terms for the European Economic Area, updated on 16 January 2026, put it in one line: “Output may not be unique and other users may receive similar output from our Services.”

GeneratorWho owns the outputTerms read
ChatGPT (OpenAI)

The user: OpenAI assigns its rights in the output, “if any”. The output may not be unique.

16 January 2026

Gemini and Google services

Google does not claim ownership of original generated content. Using it to build competing models is prohibited.

30 July 2026

Canva

The user, except where the output edits or incorporates content from the Canva library. Lawful use at your own risk.

26 June 2026

Adobe Firefly

Indemnity on eligible teams or enterprise plans, up to 10,000 dollars per output.

17 June 2025

Midjourney

Terms could not be read verbatim on 10 November 2026. Check them on the site before use, including the revenue threshold above which a higher plan is required.

Not verified

Canva's case needs a careful reading. If you edit a photo from the library with an AI tool, the terms say: “If you use an AI Product to edit a photo from the Canva library, you won’t own this Output”. The photo stays licensed, under the licence's rules. For OpenAI, the terms read are the user terms; business accounts and the API have a separate agreement, not read here.

Are images created with Gemini protected by copyright?

Google says it does not claim ownership of original content generated with its services, Gemini included; whether the image is protected by copyright is a matter for the law, not for Google. The Terms of Service in force since 30 July 2026 put it like this: “Google won’t claim ownership over that content.”

This is a waiver, neither an assignment nor a guarantee. The same rule applies as for other images: in Italy protection exists where there is human creative input, and in the United States the Copyright Office rules out prompts alone. Google's additional terms for generative AI have not applied since 22 May 2024: the general text governs today.

Faces, real people and trademarks: the limit that does not change

A person's face in an ad needs their consent, whether the image is a photo or a generation. Article 96 of Law 633/1941 is clear: “Il ritratto di una persona non può essere esposto, riprodotto o messo in commercio senza il consenso di questa”, meaning a person's portrait may not be displayed, reproduced or sold without their consent.

The exceptions in article 97, such as fame, fall away when the use harms “all’onore, alla reputazione od anche al decoro della persona ritrattata”, the honour, reputation or dignity of the person portrayed. Article 10 of the Civil Code lets a court stop the misuse of someone's image, without prejudice to damages. Whether a person's fame is enough to put their face in an advert, the law does not say plainly: for a specific case, check with a lawyer.

Since 10 October 2025 there is also a crime. Article 612-quater of the Italian Criminal Code, introduced by Law 132/2025, punishes with one to five years' imprisonment anyone spreading, without consent, images, video or voices falsified with AI and capable of deceiving, causing unjust harm. Prosecution requires a complaint from the victim.

For trademarks, article 13 of the IAP Self-Regulation Code applies: any exploitation of another company's name, trademark, reputation or corporate image intended to gain an unjustified profit must be avoided. A competitor's logo that slipped into the generated image comes out before publishing. The Commission's guidelines on article 50 of the AI Act add that the “AI-generated” label changes nothing about whether the work is protected.

Who is liable if the image infringes a right: the indemnity

The business publishing the ad is liable, unless the contract provides an indemnity, and among the terms read only Adobe Firefly provides one. Adobe's generative AI terms, in force since 17 June 2025, cover third-party claims on copyright, trademark, publicity or privacy rights for eligible Firefly outputs.

The conditions are tight. It applies only to Creative Cloud for teams or for enterprise customers on an eligible plan; outputs modified or combined with other material are excluded, as is input that already infringed; the cap is 10,000 dollars per output or per claim. The document read dates from 2025 and Adobe's site cannot be read without a browser: check the version in force before relying on it.

OpenAI, Google and Canva, in the user terms read for this page, offer no indemnity. Canva says the output is used at your own risk. For a business this means one simple thing: line 8 of the checklist, with date, prompt, original file and approver, is the only protection you can build yourself.

Italy, the European Union, the United Kingdom and the United States

The rules on who authors an AI-made image change from one country to the next, and the United Kingdom alone has a written provision for computer-generated works. The table lists the texts read on 11 November 2026, each with its date.

CountryIs an AI-made image protected?Source and date
Italy

Yes, if it is the result of a human author's intellectual work. Threshold not defined.

Art. 1, Law 633/1941, from 10 October 2025

European Union

No EU rule on protecting the output; the AI Act label does not affect protection.

Guidelines on art. 50, 2026

United States

Only the human contributions: selection, arrangement, creative modifications. Prompts alone do not count.

Copyright Office, 29 January 2025

United States, courts

A human author is required; AI assistance does not rule out protection.

Thaler v. Perlmutter, 18 March 2025

United Kingdom

For a computer-generated work, the author is the person who made the arrangements necessary for its creation.

CDPA 1988, s. 9(3)

United Kingdom, advertising

The ASA does not regulate personality rights or intellectual property, but asks advertisers to weigh the risk of using a public figure's likeness.

ASA, 11 June 2026

In the United States the Copyright Office report says: “prompts alone do not provide sufficient human control to make users of an AI system the authors of the output”. The D.C. Circuit Court of Appeals, in the Thaler case, added that the human authorship requirement does not prevent protecting works made with the assistance of AI. In the United Kingdom the ASA, the advertising self-regulator, summed it up in one sentence: “Just because a tool can generate a convincing deepfake, doesn’t mean you can safely use it.”

What remains uncertain

Three questions still have no written answer, and on these no page can give you certainty. First: how much human input is enough in Italy to protect an AI-made image. Law 132/2025 does not say, and the courts will decide case by case.

Second: whether a synthetic face resembling a real person counts as a “ritratto”, a portrait, under article 96. Third: whether someone using the output of a model trained without respecting the rights holders' reservation is liable too, alongside the company that trained it. On all three, the answer today is to check with a lawyer on the specific case.

In the United Kingdom the Government consulted in December 2024 on several points, including the provision on computer-generated works: the outcome is not verified on this page. The other parts of advertising with AI are in the guides to advertising on ChatGPT, Google Ads AI Max and what Meta Ads cost; the index is under marketing for small businesses.

Questions and answers

Can I use AI-generated images in advertising?

Generally yes: the terms of OpenAI, Google and Canva allow you to use the outputs, and no Italian law forbids it. Three limits remain: faces of real people and other companies' trademarks need consent, realistic images may need the AI Act label, and no generator read for this page guarantees that the output does not infringe anyone's rights, apart from Adobe Firefly's limited indemnity on business plans.

Can AI images be copyrighted?

In Italy, only if they are the result of a human author's intellectual work: article 1 of Law 633/1941 says so, as amended by Law 132/2025, in force since 10 October 2025. The law does not say how much human contribution is enough.

In the United States the Copyright Office states that prompts alone do not make the person writing them an author.

Are images created with Gemini protected by copyright?

In its Terms of Service, Google says it does not claim ownership of original content generated with its services. Whether an image is protected by copyright is not up to Google, though: it depends on the law of the country and on the human contribution.

Can I put an AI-generated face of a famous person in an ad?

Not without their consent. Article 96 of Law 633/1941 prohibits displaying or commercialising a person's portrait without consent, and since 10 October 2025 spreading a deepfake that causes unjust harm has been a crime in Italy (article 612-quater of the Criminal Code). For a synthetic face that merely resembles a real person the answer is uncertain: check with a lawyer.

If an AI image infringes someone's copyright, who pays?

The business publishing it, unless the contract provides an indemnity. Among the terms read on 10 November 2026, the only written indemnity is Adobe Firefly's, for Creative Cloud teams or enterprise customers on an eligible plan, capped at 10,000 dollars per output.

Can I use AI-generated images in my company’s social media posts?

Generally yes: no Italian law forbids it, and the terms of the most widely used generators allow it, in posts as well as in ads. Three questions matter: who owns the image (the generator’s contract says), whether the law protects it, and who pays if it infringes someone else’s rights.

A fourth rule applies from 2 August 2026: if the image is realistic enough to look real, Article 50 of the AI Act may require you to disclose it. Real people’s faces and other companies’ trade marks stay off limits in any case.

Notes on sources

  1. Normattiva, Law no. 633 of 22 April 1941, art. 1, art. 96 and art. 97, current text; Civil Code, art. 10. Read on 10 November 2026; the source of the quoted legal text.
  2. Gazzetta Ufficiale, Law no. 132 of 23 September 2025, general series no. 223 of 25 September 2025, in force since 10 October 2025: art. 25 (art. 1 and art. 70-septies of Law 633) and art. 26 (art. 171, letter a-ter, and art. 612-quater of the Criminal Code).
  3. Istituto dell’Autodisciplina Pubblicitaria, Self-Regulation Code, 71st edition of 30 October 2024, art. 13.
  4. European Commission, guidelines on the transparency obligations of article 50 of the AI Act, C(2026) 7702, points 127-129.
  5. OpenAI, Terms of Use for the EEA, 16 January 2026; Google, Terms of Service, in force since 30 July 2026; Canva, AI Product Terms, in force since 26 June 2026; Adobe, Generative AI Product Specific Terms, section 8, in force since 17 June 2025. Read on 10 November 2026. Midjourney's terms could not be read without a browser.
  6. U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, 29 January 2025, public domain; D.C. Circuit Court of Appeals, Thaler v. Perlmutter, 18 March 2025.
  7. United Kingdom, Copyright, Designs and Patents Act 1988, s. 9, Open Government Licence v3.0; Advertising Standards Authority, AI and Deepfakes, 11 June 2026.
·Who does it for you

An image takes a minute to generate. Finding the idea your market has not seen yet is the work that comes first. Positioning is built starting from the market.

Mattia Esposito, who wrote this guide, also works with Itria on positioning: in-depth market analysis to find where competitors are exposed and take that ground. He did it on his own site first. In August, asked who had founded itria.io, ChatGPT couldn’t find him. From 11 October to 7 November 2026 itria.io pages appeared 13,803 times in Google and Bing results and 3,818 times in AI answers: Google’s AI Overviews and AI Mode, and Microsoft Copilot.