What is an AI generated image?

From prompt to pixels

Image models learn from very large collections of pictures paired with descriptions, building an internal sense of which shapes, textures and lighting go with which words. Given a new prompt, the model starts from random noise and refines it step by step toward something that matches the description. Run the same prompt twice and the pictures differ. The strengths are photographic lighting, product shots on plain backgrounds and scenes that would be expensive to stage. The weaknesses are hands, small repeated details, exact brand marks and readable text, which models have historically rendered as convincing gibberish. None of that makes a generated picture a photograph of anything real, which matters as soon as it is used to sell something.

Who owns the result, country by country

In Australia, the Australian Copyright Council's fact sheet on artificial intelligence explains that copyright protects works with a human author, and that a work produced solely by AI falls outside it. The courts reached the same conclusion about automation in 2010, when the Full Federal Court held in Telstra v Phone Directories that directories compiled by computer systems had no copyright. The United Kingdom is the outlier on paper: section 9(3) of the Copyright, Designs and Patents Act treats the person who made the arrangements for a computer generated work as its author, with 50 years of protection, but the government's consultation of December 2024, updated on 19 March 2026, says its preference is to remove that protection if the evidence does not support it. In the United States, the Copyright Office's report of 29 January 2025 concluded that prompts alone do not make a user the author, while human selection, arrangement and modification can still be protected, and on 18 March 2025 the DC Circuit upheld the human authorship requirement in Thaler v Perlmutter.

Using an image you may not own

Lack of copyright rarely stops a business using a generated picture on a flyer, provided the terms of the tool allow it. What it removes is exclusivity. If nobody owns the image, nobody can stop a competitor copying it, so a generated picture is a weak foundation for a logo, a mascot or anything that has to identify the business. A generated image can also resemble a protected work, a real person or a real product closely enough to cause trouble, and a picture that shows a venue, dish or product that looks nothing like the real one can mislead customers regardless of who owns it. None of this is legal advice for a particular use.

Labelling and disclosure

Europe is where labelling becomes law. Article 50 of the EU AI Act requires providers of AI systems to mark generated output in a machine readable way so it can be detected as artificial, and requires anyone deploying a system to create deep fakes, meaning realistic images, audio or video of people, places or events, to disclose that the content was generated. Those duties apply from 2 August 2026. Elsewhere the obligation usually comes from advertising and consumer rules rather than an AI statute, which is why a clear, plain label on a realistic generated person or place is a sensible habit anywhere.

Common mistakes

The most common is asking the model to write the headline, price or phone number inside the picture and then printing whatever it spelled. Close behind is generating a logo and assuming the business now owns a protectable mark. Others include prompting with a real person's name or a famous character, using a generated interior as though it were the actual shop, and keeping no record of the prompt and date. That record costs nothing and is the first thing anyone will ask for if the picture is ever questioned.

How images are generated inside a flyer

In the flyer builder an image slot can be filled from a prompt of up to 2,000 characters. The picture is made by a Gemini image model on Vertex, at an aspect ratio snapped from the slot's own shape to the nearest of eight: 21:9, 16:9, 3:2, 4:3, 1:1, 3:4, 2:3 or 9:16. No text is ever rendered into the image, so prices, dates and headlines stay in text slots where they can be edited. Logo slots are never generated; they take an uploaded image or initials. Each generation needs a credit. When the generation job cannot produce a picture it tries an Unsplash stock search before leaving the placeholder, while the generate button inside the builder simply reports a failure.

Questions people ask