Yes, you can use AI images commercially, but two things surprise most business users: you probably don’t own a copyright in the image, and the legal risk for anything that image contains sits with you, not the platform. That combination shapes every practical decision, from which generator you subscribe to, to what you inspect before publishing, to what your insurance actually covers.
You Can Use the Image. You Usually Can’t Protect It.
U.S. copyright protection requires a human author. The Copyright Office’s registration guidance states that “the term ‘author,’ which is used in both the Constitution and the Copyright Act, excludes non-humans,” and that the Office “will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author.”1Federal Register. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence Writing a prompt and clicking generate does not make you the author of the result.
The commercial consequence is direct. If your hero image, product mockup, or campaign visual is purely AI-generated, a competitor can copy it and put it on their own site, and you have no infringement claim to bring. That is the trade-off attached to uncopyrightable material.
Human Edits Can Recover Some Protection
Copyright can attach to AI-assisted work when a person exercises enough creative control over the final result. The Office looks at whether “the human had creative control over the work’s expression and ‘actually formed’ the traditional elements of authorship.”2U.S. Copyright Office. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence Two routes get you there: modifying the AI output substantially enough that your changes independently meet the copyright standard, or selecting and arranging AI-generated elements in a sufficiently creative way that the finished work, as a whole, is original.
Even then, protection covers only the parts a human made. The AI-generated portions remain unprotected, and someone could extract and reuse them. You also have to disclose the AI content in the application and exclude it from your claim.1Federal Register. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence
What Your Platform’s Terms Actually Allow
Copyright law tells you what you can protect. The platform’s terms tell you what you’re allowed to do with the output, and getting that wrong can cost you your account or trigger a breach-of-contract claim on top of any legal issue with the image itself.
Midjourney
Midjourney gives subscribers ownership of the images they create and the right to use them commercially, with one revenue trigger: if your business grosses more than $1,000,000 per year, commercial use requires a Pro or Mega plan.3Midjourney Docs. Using Images and Videos Commercially If you upscale an image another user created, it belongs to that original creator, and you’d need their permission to use it.
OpenAI (DALL-E, ChatGPT image tools)
OpenAI’s terms assign OpenAI’s right, title, and interest in outputs to you. That sounds sweeping, but because other users can receive similar or identical output, that assignment does not give you exclusive rights. OpenAI also prohibits representing AI-generated output as human-made and bars using output about a person for decisions with legal or material impact, such as employment or credit.4OpenAI. Terms of Use
Stability AI (Stable Diffusion)
Stability AI’s community license permits free commercial use for individuals and organizations generating under $1,000,000 in annual revenue. Above that line, you need an enterprise license. You own the outputs, but you can’t use them to train competing foundational models.5Stability AI. Stability AI License
Indemnification Is Not a Given
Some platforms indemnify you if a third party claims your AI output infringes their intellectual property. Google Cloud offers a two-pronged indemnity covering both training data and generated output allegations, though the output protection applies only if you didn’t intentionally try to create infringing content.6Google Cloud Blog. Protecting Customers with Generative AI Indemnification Other platforms offer nothing. Read that clause before you build a campaign on any given service.
Trademarks Inside the Image
Models trained on huge image sets will sometimes generate output containing recognizable logos, brand marks, or product designs. If you publish that commercially, you can face a Lanham Act trademark claim. The core test is whether consumers are likely to be confused about the source of your goods or services, or to believe the trademark owner sponsors or authorized your use.7Marquette Intellectual Property and Innovation Law Review. AI, The New Frontier: An Analysis on Trademark Litigation Strategies in the Face of Generative Artificial Intelligence
The remedies are heavy. A trademark owner can recover the defendant’s profits, actual damages, and costs, and courts can multiply damages up to three times the actual amount. For intentional counterfeit-mark cases, statutory damages run from $1,000 to $200,000 per counterfeit mark per type of goods or services, rising to $2,000,000 for willful infringement. Attorneys’ fees are available in exceptional cases.8Office of the Law Revision Counsel. 15 USC 1117 – Recovery for Violation of Rights
Getty’s UK litigation against Stability AI shows how this plays out in practice. The High Court found limited trademark infringement where earlier versions of Stable Diffusion replicated Getty’s watermark in generated outputs.9Mayer Brown. Getty Images v Stability AI: What the High Court’s Decision Means for Rights Holders and AI Developers Inspect every image for brand elements, watermark artifacts, and anything that looks like it belongs to someone else before it goes out.
Images of Real People
Generators can produce images that resemble identifiable individuals, and commercial use of those without consent creates exposure under right-of-publicity laws. Most states protect a person’s ability to control the commercial use of their name, image, and likeness, and the specific protections and remedies vary considerably.
Federal legislation has been proposed. The NO FAKES Act would create a federal intellectual property right in every individual’s voice and likeness, extend that right to families after death, and give individuals a cause of action against anyone who knowingly creates, posts, or profits from unauthorized digital copies.10Representative Maria Salazar. Congresswoman Salazar Introduces the NO FAKES Act As of early 2025 the bill remains in committee and is not law.11Congress.gov. S.1367 – NO FAKES Act of 2025 Until that changes, you’re navigating the state-by-state patchwork, and the safest path is to avoid commercial use of any AI image depicting a recognizable person without a release.
Labeling and Disclosure
The FTC’s position is that “there is no AI exemption from the laws on the books,” and the agency has brought enforcement actions against companies using AI to generate deceptive content.12FTC. FTC Announces Crackdown on Deceptive AI Claims and Schemes There is no blanket federal rule requiring every AI commercial image to carry a label, but using AI images in ways that mislead consumers falls squarely within existing prohibitions on deceptive practices. If your AI image makes a product look different from what customers receive, or implies an endorsement that isn’t there, you’re exposed.
The EU AI Act goes further. Article 50, effective August 2, 2026, requires providers of AI systems that generate images, video, or audio to ensure outputs are “marked in a machine-readable format and detectable as artificially generated or manipulated.” Deployers of AI that generates deepfakes must disclose that the content has been artificially generated or manipulated.13Artificial Intelligence Act. Article 50 – Transparency Obligations for Providers and Deployers of Certain AI Systems If your AI images reach European consumers, these obligations apply to you.
OpenAI’s terms already prohibit representing output as human-generated.4OpenAI. Terms of Use Labeling AI commercial content proactively costs little and reduces deceptive-practices exposure ahead of rules that are clearly coming.
Training-Data Lawsuits
The unresolved question hanging over every AI image is whether the underlying model was trained on copyrighted material without authorization. The U.S. Copyright Office’s 2025 report on generative AI training acknowledges that when a model can generate output substantially similar to training examples, that expression “must exist in some form in the model’s weights,” which could constitute prima facie infringement.14U.S. Copyright Office. Copyright and Artificial Intelligence, Part 3: Generative AI Training The report does not resolve whether end users face downstream liability for using outputs from a model trained on infringing material.
In the Getty case, the UK High Court ruled that AI model weights are not themselves a “copy” of the training images in the copyright sense, and rejected the central copyright claim.9Mayer Brown. Getty Images v Stability AI: What the High Court’s Decision Means for Rights Holders and AI Developers That’s one ruling in one jurisdiction, and U.S. courts may reach different conclusions. The practical response is to use platforms that offer IP indemnification and avoid prompts that name specific copyrighted works, artists, or styles.
Check Your Insurance Before You Need It
Standard errors and omissions coverage often restricts protection to failures of software your organization developed, which can leave you uncovered when a third-party AI product malfunctions and triggers litigation against you.15Harvard Law School Forum on Corporate Governance. The Hidden C-Suite Risk of AI Failures Some carriers are adding explicit AI exclusions to E&O, D&O, and cyber liability policies.
At each renewal, review your policies for AI-related exclusions and ask your carrier whether AI content claims are covered. If you find an exclusion, request its removal or move to a carrier that offers explicit coverage.
Before You Publish
The risks above collapse into a set of steps to run through before any AI image goes out commercially:
- Inspect for recognizable logos, watermarks, brand elements, and real people’s likenesses. Generators frequently produce plausible-looking brand marks close enough to real trademarks to create confusion.
- Check for the visual tells of AI generation, including malformed hands, inconsistent lighting, garbled text, and impossible physics. These don’t create legal liability but they undermine credibility.
- Confirm your subscription tier covers commercial use. If your business exceeds the revenue thresholds Midjourney and Stability AI impose, upgrade before publishing.
- Save the prompt, the platform, the date, and the subscription tier. If a claim arises, you’ll want to show what you asked for and where.
- Consider embedding provenance metadata. The C2PA standard cryptographically binds provenance information to images so how content was created and modified can be verified, and tools built on it can sign assets and embed persistent watermarks.16Adobe for Business. Content Authenticity Arrives for Enterprises
- Label AI-generated commercial content even where not yet legally required in your jurisdiction. It’s cheaper than defending a deceptive-practices claim.
None of this eliminates risk. The legal framework around AI images is being built in real time through agency guidance, court decisions, and pending legislation. What protects you now is choosing a platform with clear commercial terms and indemnification, inspecting every image before it goes out, and keeping documentation that shows you acted reasonably.