Generating a product photo, an ad visual, or a first-pass logo concept with a tool like Midjourney, DALL-E, or Adobe Firefly has become an ordinary part of how small businesses produce marketing material — a usable image in minutes, for a fraction of a photographer's or designer's fee. What most businesses making that trade haven't worked through is that it carries two separate legal problems, pulling in opposite directions. The first is that the image may not belong to the business in any protectable sense. The second is that the same image may infringe on something that belongs to someone else. A single file can carry both risks at once.
The ownership problem starts with a plain rule the U.S. Copyright Office has held to consistently: copyright protects human authorship, and an image produced by typing a prompt and accepting what the model returns generally has none of the human creative choice-making the law requires. The Supreme Court left that rule in place in March 2026 when it declined to hear Thaler v. Perlmutter, closing off the leading attempt to argue that purely AI-generated output should qualify for copyright on its own. The consequence is easy to miss at the moment the image is created, since it looks like a finished ad rather than an unprotected one. A competitor who copies that exact image has not necessarily infringed anything, because there was no copyright there to infringe.
The infringement problem runs the other way, and it's the one that actually gets a business sued. Image-generation models were trained on vast volumes of copyrighted and trademarked material, and can produce output that echoes it closely enough to matter — a logo reading as a real brand's mark, a face resembling an identifiable celebrity, a character close enough to a franchise property that the resemblance isn't coincidental. Disney and Universal's June 2025 lawsuit against Midjourney, built around the platform's tendency to generate recognizable Marvel and Star Wars characters on request, is the clearest public example of how literally that risk plays out. A business that publishes a generated image containing someone else's trademark or likeness is exposed the same way it would be if a designer had copied that material on purpose. Intent isn't the test in trademark and publicity-rights claims; the commercial use is.
What makes this combination easy to walk into is a false sense of clearance from the tool itself. Most AI image platforms grant broad commercial-use rights in their terms of service, and it's a reasonable-sounding assumption that this means the output is safe to publish. It doesn't, because a vendor can only grant rights it actually holds. It cannot waive a third party's trademark, license someone's face, or manufacture a copyright claim where the law says none exists. The terms describe what the vendor owes the business — nothing more. What a third party can claim against the business for what it published is a separate question entirely.
The exposure isn't uniform across every generated image. A one-off social post, generated and used once, carries limited risk on either front — there's little to lose by not owning it, and a generic scene is unlikely to contain anyone's protected mark. The risk rises sharply for anything meant to be durable: a logo, a mascot, packaging art, or a signature visual built into a paid campaign that runs for months. Those are exactly the assets a business most wants to own outright, and may not. The infringement risk rises separately whenever a prompt names a real brand, a public figure, or a named artist's style — "in the style of," phrased that directly, increases rather than reduces the odds of a recognizable, actionable result.
The fix doesn't require abandoning these tools, which remain a genuinely useful way to produce ordinary marketing material. It requires sorting generated images into two buckets before they go live. Disposable content can stay purely AI-generated, on the understanding that it isn't a protectable asset and shouldn't be treated as one. Anything meant to become a durable brand identifier needs a human designer to meaningfully revise, combine, or build on the AI draft, with that contribution documented, since deliberate human authorship is what creates a real copyright claim. And before anything goes to press, someone should check it specifically for a recognizable trademark, a real face, or a named artist's signature style a prompt may have pulled in — a five-minute check that catches the failure mode Disney's lawsuit made public.
None of this is a reason to slow down using AI to produce marketing material, which is doing real, legitimate work for businesses that never had the budget for a full creative team. It's a reason to stop treating "AI-generated" and "safe to build a brand on" as the same thing. The businesses that avoid trouble here won't be the ones that stopped generating images. They'll be the ones that knew which images were disposable, which were meant to last, and put a human hand on the ones in the second category before they became the business's actual brand.
- ai image generation
- copyright
- trademark
- marketing
- intellectual property