Made images with ChatGPT, Gemini or Midjourney? What you can legally do

Who owns an image you made with ChatGPT or Midjourney?

There are two layers. The first is the contract with the tool provider. OpenAI’s terms say that, as between you and OpenAI, you own the output and OpenAI assigns you whatever rights it has. Google’s terms say it will not claim ownership of content you generate. Midjourney says paying users own their images, with conditions covered below.

The second layer is copyright, which is a right against the whole world. Here the answer is less comfortable. Most legal systems protect only work that reflects human creative choices, so a purely machine-made picture may not be protected at all. The tool can only hand you rights it has, and if no copyright exists, there is nothing to hand over.

In practice, this means you can usually use the image, sell products with it and put it in your marketing. What you may not be able to do is stop a competitor from copying it. That matters most for logos, mascots, book covers and anything you expect to defend.

Can you copyright an AI image? The position country by country

The US is the clearest. In its January 2025 report on copyrightability, the US Copyright Office said prompts alone generally do not make you the author, because you do not control how the system turns them into an image. Human selection, arrangement and real modification can be protected. In the Zarya of the Dawn decision (2023), the Office registered a comic book’s text and the author’s arrangement, but not the Midjourney images inside it.

The courts agree. In Thaler v. Perlmutter the D.C. Circuit held in March 2025 that a work made autonomously by an AI system cannot be registered, and the Supreme Court declined to hear the case on 2 March 2026. The EU test of originality, the author’s “own intellectual creation”, points the same way, although no top EU court has ruled on AI images yet.

Other markets differ. The UK’s 1988 Act protects “computer-generated works” with no human author, giving ownership to the person who made the arrangements for the work and 50 years of protection. In March 2026 the UK government recommended removing that rule, but no law has changed yet. China’s Beijing Internet Court in Li v. Liu (2023) protected a Stable Diffusion image because of the user’s detailed prompts and adjustments. India’s Copyright Act also names the person who causes a computer-generated work to be created as its author, but how that applies to modern AI is untested in the courts.

Can you trademark a logo made with AI?

Often, yes. A trade mark is a sign that tells customers which business a product or service comes from. Trade mark offices ask whether the sign is distinctive and not confusingly similar to existing marks. They do not generally ask who, or what, drew it.

So an AI-made logo can be registered as a trade mark in the US, UK, EU, India and most other markets, even if it has no copyright. Registration then gives you rights against anyone using a confusingly similar sign for similar goods or services. That is often the protection a small business really needs.

The catch is clearance. AI tools can produce logos that look like existing brands, because they learned from millions of them. Search the trade mark registers in your markets before you print anything, and add your own human design changes so you also have some copyright in the final version.

Can an AI image infringe someone else’s copyright or trade mark?

Yes. If the output reproduces a recognisable part of a protected work, publishing it can infringe copyright, whoever typed the prompt. Famous characters are the obvious risk: a picture of Mickey Mouse, Spider-Man or Pikachu on your product is a problem even if a machine drew it. Brand logos and packaging that appear in an image can also raise trade mark issues.

Style is different. Copyright protects expression, not a general style or idea. That is why the “Ghibli-style” portraits that went viral after ChatGPT added native image generation in March 2025 are generally not infringing in themselves. The risk rises when the image copies specific characters, scenes or frames, or when you market it as connected to the studio.

Courts are now testing where the lines sit. Disney and Universal sued Midjourney in Los Angeles in June 2025, claiming it trained on and generates their characters such as Darth Vader and the Minions, and Warner Bros. Discovery filed a similar case later that year. As of October 2026 those cases are still pending. In the UK, Getty Images largely lost its case against Stability AI in November 2025, but the court found trade mark infringement where early versions of Stable Diffusion produced images bearing Getty and iStock watermarks.

• Avoid prompts naming famous characters, brands or living artists for commercial work.
• Check outputs for stray logos, signatures and watermarks before publishing.
• Run a reverse image search on anything important, such as a cover or logo.

Can you use a real person’s face in an AI image?

This is often the biggest risk in practice. Many countries protect a person’s name, face and voice against commercial use without consent. In the US this is called the “right of publicity” and is mostly state law; California and New York have strong versions. In India the courts call it “personality rights”, and in September 2025 the Delhi High Court granted interim orders protecting Aishwarya Rai Bachchan and Abhishek Bachchan against AI deepfakes and misuse of their likeness.

The UK has no general image right, but using a celebrity in an ad can still be “passing off”, meaning a false suggestion that they endorse you. Advertising codes in the UK, EU and elsewhere also forbid implying an endorsement that does not exist. Data protection law adds another layer: an identifiable face is personal data under the GDPR and similar laws.

Made-up people are not risk-free either. Since June 2026, New York requires advertisers to disclose conspicuously when an ad uses a “synthetic performer”, an AI-generated human who is not a real, identifiable person. Penalties are $1,000 for a first breach and $5,000 for later ones.

What the tool’s terms of service say about commercial use

Terms matter more than most people realise, and they change. OpenAI and Google give you their rights in the output and allow commercial use within their usage policies. Midjourney says paid subscribers own their images, but a company with more than $1 million in annual gross revenue, or its employees, must be on a Pro or Mega plan to own them. Midjourney images may also be visible to other users unless you use its private mode.

Adobe markets Firefly as “commercially safe” because it says the models were trained on licensed content such as Adobe Stock and public domain works. Qualifying enterprise plans include an intellectual property indemnity, a promise that Adobe will defend you and cover certain costs if a generated image is said to infringe. Indemnities usually have conditions and do not cover your own edits, so read them closely.

Free and consumer plans rarely offer any indemnity. Stable Diffusion models come under their own licences, which can restrict commercial use above a revenue level depending on the model version. Screenshot the terms that applied when you made the image and keep them with the file.

Do you have to label AI images?

Increasingly, yes, in certain cases. The EU AI Act’s Article 50 requires providers of generative tools to mark outputs in a machine-readable way, and requires businesses using them to disclose “deepfakes”, meaning realistic images of real people, places or events. These duties have applied since 2 August 2026, and fines can reach €15 million or 3% of worldwide turnover, with lower caps for small companies.

China’s labelling measures, in force since 1 September 2025, require visible labels and hidden metadata on AI content. India’s amended IT Rules, in force since 20 February 2026, require platforms to label “synthetically generated information” and keep embedded markers. California’s AI Transparency Act puts watermarking duties on large AI providers.

Many tools now mark images automatically. OpenAI and Adobe attach C2PA “Content Credentials”, a signed record of how the file was made, and Google adds its invisible SynthID watermark to images made with its models. Do not strip these markers. Removing them can breach platform rules and, in some markets, the law.

Advertising rules: when an AI image misleads customers

Advertising law does not care how an image was made. It asks whether the ad gives a false impression. A product photo that shows a dish, room or garment better than the real thing can mislead, whether it came from a camera, Photoshop or Gemini. In the US the Federal Trade Commission (FTC) polices deceptive ads, and its 2024 rule bans fake reviews, including AI-generated ones.

The UK’s Advertising Standards Authority (ASA) has said ads made with AI are subject to the same rules as any other ad. Its earlier rulings against beauty ads that used filters to exaggerate a product’s effect show how it will likely treat AI-enhanced product images. EU consumer law and India’s Consumer Protection Act take the same approach to misleading ads.

The safest practice is to use AI for mood and illustration, and real photographs for what the customer will actually receive. Where an AI image could be mistaken for a real product shot or a real customer, say so.

The market now: an image boom, and the courts catching up

AI image making went mainstream in 2025. ChatGPT’s native image generation in March 2025 set off the Ghibli-style trend within days, and Google’s Gemini image models drew a similar wave of viral edits later in the year. Businesses quickly moved from experiments to everyday marketing, product listings and client work.

Rights holders responded. The studio lawsuits against Midjourney, Getty’s case against Stability AI and many other claims are testing whether training and outputs infringe. Stock libraries took different paths: Getty Images and Shutterstock have generally not accepted AI-generated uploads from contributors, while Adobe Stock accepts them if they are labelled and do not use third-party content without permission.

The direction of travel is fairly clear. Courts are unlikely to give copyright to purely machine-made images, but will protect real human input. Liability for outputs that copy characters or faces is growing. And labelling is moving from good practice to legal duty. If you treat AI images like stock photos you do not fully own, you will usually stay on the right side of these trends.