AI image and video tools for marketing: copyright, disclosure and brand risk

AI image and video tools can produce a lifestyle shot, a product background or a short social clip in minutes, for a fraction of the cost of a shoot. For small brands the saving matters. It also raises questions that a traditional photoshoot never did: who owns the result, whether it copies someone else’s work, whether you need to tell people it is AI, and what happens to trust if customers feel misled.

This guide sets out the UK position as of October 2026, with a note on EU rules for brands that sell there. It is general information, not legal advice. Several of these areas are under active review, so check the current position before a major campaign.

Who owns an AI-generated image in the UK

UK copyright law has an unusual provision for “computer-generated” works with no human author. Under section 9(3) of the Copyright, Designs and Patents Act 1988, the author is taken to be “the person by whom the arrangements necessary for the creation of the work are undertaken.” That section is still in force.

Its future is uncertain. The government’s Report on Copyright and Artificial Intelligence, published in March 2026, says protection for wholly computer-generated works “should be removed”, while works created with AI assistance would continue to be protected. In practice:

• Do not count on owning strong rights in an image produced from a single prompt.
• The more human creative input you add, through composition, editing, combining with your own photography and design choices, the stronger your position is likely to be.
• Read the tool’s terms on ownership and commercial use. They differ between providers and between free and paid plans.

Could the output copy someone else’s work

The same report notes that an AI service “may generate a substantial part of one or more copyright works in its output”, which without a licence could infringe copyright. It also refers to the Getty Images v Stability AI case, in which a trained model was held not to be an infringing copy because there was no evidence it contained copies of the works. That point was under appeal when the report was written.

For a marketer, the practical risk is the output, not the model. Reduce it by:

• not prompting with living artists’ names, other brands, film characters or logos;
• running a reverse image search on any image you plan to use prominently;
• keeping a record of the prompts, tool, date and edits for each asset.

Where UK policy on training and labelling stands

The March 2026 report dropped the government’s earlier preferred option of a broad text and data mining exception with an opt-out for rights holders. It did not require AI developers to disclose their training material, choosing instead to develop best practice on transparency with industry. On labelling AI-generated content, it proposed no legal obligation and said the government would work with industry on best practice and watch international developments.

So for now there is no general UK law requiring you to label AI images. Advertising rules still apply, and they are where most of the day-to-day risk sits.

Advertising rules: what the ASA says

In a May 2025 article on AI disclosure, the Advertising Standards Authority said there is no blanket legal requirement to disclose AI use in ads, and the advertising codes have no AI-specific rules. The existing rules apply “regardless of how content is generated, edited, or targeted.”

Three points stand out:

1. Disclosure does not fix a misleading ad. The ASA says “disclosure alone is very unlikely to mitigate the harm caused by a fundamentally misleading message.” Its example is an AI image showing a cosmetic effect the product does not deliver, which is likely to be materially misleading even with a disclosure.
2. Disclosure can help in some cases. Making clear that an influencer is AI-generated “could well help to negate an otherwise misleading impression”, depending on context.
3. You are responsible for what automated tools produce. A June 2026 summary by Lewis Silkin describes ASA rulings in which advertisers were held responsible for ads made with automated tools, and one against an AI video app whose ad showed intimate deepfakes of real celebrities.

Real people, likeness and deepfakes

Using AI to depict a real person, especially a celebrity, carries the highest risk. It can suggest an endorsement that does not exist, which is misleading, and it can raise passing off and other legal issues outside the ASA’s remit. The March 2026 report says the government will consider “whether it would be beneficial to introduce a new digital replica or personality right.”

If you sell into the EU, Article 50(4) of the EU AI Act requires deployers to disclose when image, audio or video content is a deepfake, with a lighter duty for evidently artistic, creative or satirical work. Article 50(2) separately requires providers of generative tools to mark outputs in a machine-readable way. According to the Future of Privacy Forum’s AI Act timeline, the transparency rules apply from 2 August 2026, with a grace period to 2 December 2026 for marking by tools already on the market before that date.

Brand risk beyond the law

Staying legal is the minimum. AI imagery can also drift off-brand, repeat stereotypes and look generic next to competitors using the same tools. In the celebrity deepfake ruling, the ASA also told the advertiser to avoid harmful gender stereotypes in future. Customers who feel tricked may lose trust even when no rule was broken.

Set simple house rules: which uses are fine (backgrounds, mood boards, concept sketches), which need sign-off (people, lifestyle scenes), and which are off limits (product results, real people without written consent, fake customers or reviews).

A worked example and common mistakes

An example (illustration only): a small skincare brand wants AI lifestyle images for a launch.

1. It uses AI for bathroom and shelf backgrounds, then places real photos of the product into them.
2. It does not use AI to show skin before and after, because that would depict a result the product may not deliver.
3. It avoids generating faces that resemble known people and uses a clearly fictional model only for mood images.
4. It keeps a log of prompts, tools and edits, and runs a reverse image search on each hero image.
5. For its EU site, it adds a short note where an image of a person is AI-generated.

Common mistakes include relying on a disclosure to rescue a misleading image, assuming AI output is automatically yours to protect, using a celebrity lookalike, and skipping a record of how each asset was made.