AI images: licensing and commercial use

Generating an image takes seconds. Deciding whether you may use it in a paid campaign is far harder. Three layers overlap: what the tool’s terms say, what copyright law says, and who can sue you. This piece lays out the 2026 reality for marketers, founders and designers. Stated plainly up front: this is general information, not legal advice. Any specific case — especially a large campaign, product packaging or a dispute — needs a lawyer who knows your jurisdiction.

Three questions to keep separate

Most confusion comes from blending three distinct issues. Separate them and the picture clears:

• May I use this? – answered by the tool’s terms (the licence).
• Do I own it exclusively? – answered by copyright law.
• Can someone sue me? – third-party risk: training-data creators, trademark owners, people whose likeness appears.

A tool can grant you commercial-use rights (question one) while you still cannot stop anyone copying the image (question two) and still carry litigation risk (question three). Three different axes.

Who owns the AI output

Here is the trap that surprises most people. The US Copyright Office has consistently held that copyright protection requires human authorship. A prompt alone, however elaborate, does not under their guidance make you the author of the generated image — because the model, not you, makes the creative choices about composition, colour and detail.

Practical consequence: a purely generated image in the US may not be covered by copyright at all. You then cannot stop a competitor using an identical shot, and a registration attempt may be refused or limited to your human edits. The more genuine human work (retouching, composition, combining with other elements, selection and arrangement), the better the odds of protection — but only over that human layer. In the EU the statutory wording can differ, yet the common denominator — human creativity as a condition of protection — is similar. If you need exclusivity (a logo, a flagship key visual), do not base it on raw model output.

What the tool terms say

“I may use it” is a licensing question, and licences differ significantly. General state of play in 2026:

• Midjourney – commercial-use rights are typically tied to an active paid plan. On free access (when open) commercial use can be excluded, and large companies face separate thresholds. Images are public in the gallery by default unless you buy a private mode.
• OpenAI (DALL·E, image generation) – the terms generally assign generated content to the user and permit commercial use, subject to the usage policies (no prohibited content, no impersonation).
• Adobe Firefly – positioned as “commercially safe”, trained on sources such as licensed Adobe Stock and public-domain material. For some plans (business ones in particular) Adobe offers indemnification – a commitment to cover certain legal claims – but always check the scope and conditions for your plan.
• Stable Diffusion / Flux and open models – here “open” does not mean “do whatever you like”. Specific licences apply (e.g. the OpenRAIL family with acceptable-use clauses, and with some providers separate commercial licences or limits for high revenue). Community and enterprise builds can carry different terms.

Practical rule: read the terms on the day you use them, for your specific plan, and archive a copy. Terms change, and in a dispute what counts is what applied when you created the material.

Training-data lawsuits and your risk

A wave of litigation is under way in which creators and rights holders accuse AI companies of unlawfully using their works to train models. I will not resolve them here — they are ongoing and look different across countries. Two things matter for you as a user:

1. These cases mostly target model providers, but that does not make the user fully safe. If a generated image too closely reproduces a specific protected work, or a recognisable style tied to a living artist, your commercial use can also become the subject of a claim.
2. Provider indemnification (as in some Adobe plans or certain enterprise offers) shifts part of that risk — but usually on conditions: you use it within the terms, you do not bypass filters, you do not feed someone else’s protected content as input.

My view, clearly flagged as opinion: for high-risk campaigns (paid media, print, packaging, a brand’s face) choose tools with explicit indemnification and clean data provenance, and keep experimental models for internal sketches.

Trademarks and likeness – the quiet trap

Even an image that is perfectly “clean” on copyright can fall into a different trap. Copyright is not everything:

• Trademarks – logos, distinctive packaging, a product shape protected as a design. A model can “paint in” a recognisable logo on a shoe or a can. Using that in an ad risks infringement.
• Likeness and image rights – a face resembling a specific, recognisable person (a celebrity, but also an ordinary individual) can violate personality rights or the right to commercial use of likeness.
• Characters and others’ property – well-known characters from films, games and cartoons are often protected in layers. “It is only AI” is not a defence.

Rule of thumb: if the image contains something you recognise as someone else’s brand, face or character — treat it as a red flag and do not publish it commercially without checking.

When you need a model release

A model release is a person’s consent to commercial use of their likeness. For purely synthetic images, where the face is no real person’s face, a classic release has no one to apply to. But watch three situations:

• You used a real person’s photo as input (a reference, img2img, fine-tuning) – then you need their consent as with ordinary photography.
• The output “accidentally” resembles a specific, recognisable person – the risk of a claim exists despite the synthetic origin.
• You combine AI with real material (a model’s face, a location, a private interior, an artwork in the background) – standard consents and property releases come back into play.

Transparency duties (AI Act)

In the EU, the AI Act introduces transparency duties for content generated or substantially modified by AI — in simple terms: such content should be possible to recognise as artificially produced (for example through labelling, including machine-readable marks). This is not a ban on using AI images in marketing. It is a requirement not to pass them off as authentic photography where that misleads the audience. The timeline and implementation details are worth tracking — and again: this is a general description of direction, not an interpretation. Independently of the AI Act, unfair-commercial-practices rules already penalise misleading advertising today.

Checklist before using an AI image in a paid campaign

1. Check the tool’s terms for your plan, as of today, and save a copy.
2. Make sure the plan is paid where commercial use requires it (e.g. Midjourney).
3. You did not feed someone else’s protected photos as input, nor “in the style of” a specific living artist.
4. The image contains no third-party logos, marks, packaging or recognisable characters.
5. No face resembles a specific, recognisable person (and if you used a real person – you hold a model release).
6. For high-risk campaigns: pick a tool with indemnification and clean data provenance.
7. In the EU: label the content as AI-generated where it would otherwise mislead.
8. Do not base exclusivity on raw output — add a genuine human layer if you want protection.
9. When the stakes exceed the cost of advice — ask a lawyer. It is cheaper than a dispute.

TL;DR

Separate three questions: may I use it (the tool’s licence), do I own it exclusively (copyright — in the US raw AI output is often unprotected), can someone sue me (training data, trademarks, likeness). Read the terms for your plan and archive them. For high-risk campaigns choose tools with indemnification. Watch for logos, faces and characters even in “synthetic” images. In the EU, label AI content where it would otherwise mislead. And remember: this is general information, not legal advice — when real money is on the line, consult a lawyer.