September 5, 2026
Can you sell an AI-generated image? Rights, licences and commercial use
What you actually own on an AI image or video, why ownership and copyright are two different things, and the four situations where you take a real legal risk.
Frank HoubreFounder of Imaginode
The question that stops everyone before the first euro is spent
You can use your generations commercially in nearly every case, but the legal protection of those images against copying is weak, and the two points have nothing to do with each other.
This is the question that comes up most often when I work with teams. Before we even talk about models or prompts, someone raises a hand: am I allowed to sell this? To put it on packaging? To hand it to a client who will broadcast it on television?
The short answer is yes, in the overwhelming majority of cases. But the question hides two very distinct questions that almost everyone mixes together, and the mix is expensive. First question: am I allowed to use this image commercially? Second question: am I protected if someone copies it? The two answers are not the same.
We are going to untangle that properly, with what the service terms actually say, what copyright says, and above all the four concrete situations where you take a real risk. One useful note before we start: I am a filmmaker and a trainer, not a lawyer. What follows is what I apply in production, not legal advice. For a contract with serious money on it, have a lawyer read it.
What Imaginode's terms say in black and white
Content generated by users belongs to them, within the limits of the licences of the model providers used: the platform claims no right over your images.
Let us start with the easiest thing to check, because it is written down and enforceable. The terms of service say that content generated by users through the service belongs to them, within the limits of the licences of the model providers used. Two pieces of information in a single sentence.
The first: Imaginode keeps nothing for itself. No exploitation licence on your visuals, no right to republish them, no clause allowing them to be put in a showcase. Your projects live in private storage, and what you generate is yours. Check that point on every platform you use, it is not universal.
The second piece of information is the one people skip: within the limits of the providers' licences. When you generate with Flux 2 Pro or Seedream 5 Lite, it is the provider's model that produces the image, and its own licence applies on top. We come back to that below, because that is where the rare bad surprises hide.
Ownership and copyright: the confusion that costs money
Owning a file and holding copyright over it are two separate things: you can perfectly well exploit an image without holding any legal monopoly on it.
Here is the distinction nobody makes, and it explains almost every muddled discussion on the subject. Owning an image means the file is yours, that you can sell it, print it, distribute it, invoice it. That is ownership in the ordinary sense, like a chair.
Copyright is something else: a legal monopoly that lets you stop others from copying. It is not bought and not declared, it arises automatically when a work carries the imprint of its author's personality. It is that human imprint that becomes a problem with an automatic generation.
Keep in mind that the two are independent. You may well have the right to use an image commercially without having the right to stop a competitor from reusing it. That is uncomfortable, but it is the current situation, and it has very concrete consequences on what you put in your deliverables.
Why a raw generation is poorly protected
Copyright protection assumes identifiable human creative choices: an image produced from a prompt in one click struggles to tick that box, in France as in the United States.
French copyright protects original works, originality being understood as the imprint of the author's personality. A court looks for creative choices: framing, light, composition, what the person decided rather than accepted. Faced with an image produced from a twelve word prompt in one click, those choices are thin.
In the United States, the copyright office holds a public and consistent position: registration requires a human author, and an image produced mechanically by a model does not meet that condition. Applications have been refused on that basis. The doctrine moves, decisions come out regularly, but the general direction has been stable for several years.
It is not the end of the world, granted. But concretely, if you sell an illustration generated in thirty seconds and a competitor reuses it, your position for going after them is weak. And the more your work is limited to typing a prompt, the weaker it gets.
How to take back control over the protection of your visuals
The more documented human work your production contains, editing, retouching, art direction, iterations, the better the whole thing holds up: it is your chain of decisions that becomes protectable.
The good news is that you control that dial. A visual that comes out of twenty iterations, a held art direction, a reframe, colour work and an assembly with your own elements, is no longer a raw generation. It is a composition in which your choices are everywhere.
That is exactly what the node canvas documents along the way, with nothing extra to do on your side. Your successive attempts stay in each node's history, your art direction lives in a Style node, your framing settings in the Camera node. The day someone asks you to prove your work, you have the full path.
Personally, on anything that goes to a client and has to last, I keep the whole project. Not just the final file: the canvas, the discarded versions, the dates. It costs nothing to keep and it turns a painful discussion into a short one.
Risk situation number one: the style of a living artist
Naming a contemporary artist in a prompt to copy their style exposes you, especially if the result is recognisable and distributed commercially.
First real danger zone, and by far the most common. You type the name of a living illustrator into your prompt because you like their line, the model gives you something very close, and you put it on a paid campaign.
Style in itself is not protected, that is true, and many people stop at that sentence to reassure themselves. But if the result is close enough to identifiable works, we are no longer talking about style, we are talking about reproduction. And the artist's name written in your prompt is proof of intent served on a platter.
Never put the name of a living artist in a commercial prompt. Describe what you like about them: the texture, the palette, the line, the light, the grain of the paper, whatever it is. You get the look without the name, and it is also a much better way to prompt, because a model understands technical descriptions better than a signature.
Risk situation number two: trademarks and characters
A registered trademark or a protected character stays protected whatever tool was used to draw it: AI launders nothing.
Second red zone, more obvious but massively ignored. A brand logo, a recognisable bottle, a studio character, a game cover. The fact that a model generated them changes strictly nothing about their status.
Models saw those elements during training and reproduce them very well. You ask for a generic can, it gives you something that looks a lot like an existing brand. Look at your images before delivering them, literally: labels, bottle shapes, logos on clothes, posters in the background.
It is dumb, plain proofreading work, and it is the one people skip because the image is pretty. On a client production, I always do one pass where I look at nothing but the backgrounds. The Enhance node also lets you zoom into details before printing, and that is often where the surprises turn up.
Risk situation number three: a real person's face
Personality rights protect real people independently of copyright: generating or reconstructing an identifiable face without consent exposes you, even if the image is entirely synthetic.
Third trap, and the one rising fastest with reference tools. You load three photos of a person into a Reference node, and the model reproduces them very faithfully from one image to the next. Technically it is superb. Legally it depends entirely on who that person is and what they signed.
Personality rights are a right of the person, not a copyright. They exist even when the image is synthetic, as long as the person is identifiable. A celebrity, a former client, a colleague whose LinkedIn portrait you reused: without written consent, you have no business doing anything commercial with that face.
For a professional portrait or a virtual try-on on yourself or on a model who signed, no problem at all. For an entirely invented face, no problem either, and that is often the right answer: ask the model for a fictional person rather than starting from a photo you found.
Risk situation number four: the model's own licence
Some providers restrict commercial use of their model, especially on free tiers: that is the limit Imaginode's terms explicitly point to.
Back to the sentence in the terms: within the limits of the licences of the model providers used. Every model in the catalogue comes with its own rules, and they do not all say the same thing.
The rule that comes up most often across providers, whatever the tool, is the distinction between the free offer and the paid one. Many platforms allow commercial use only to paying subscribers and leave free generations for personal use. That is one of the real hidden costs of free, which I cover in detail in what a free AI really costs.
On Imaginode, you generate with a catalogue of commercial models paid per use, and each model page tells you what it does. If a project involves serious budgets, take five minutes to open the licence of the provider whose model you plan to use for the final version. Five minutes, once, on a model you will keep for months.
The special case of the logo, which escapes the rule
A logo is protected by trademark registration, a mechanism entirely independent of copyright: an AI-generated identity is registered and defended normally.
The logo deserves its own paragraph because it worries everyone and is in fact the most comfortable case. Protecting a logo goes in practice through trademark registration, not through copyright.
A trademark is registered with an office, for classes of goods and services, and what counts is priority and distinctiveness, not how the drawing was produced. A logo out of an AI logo generator is registered exactly like a hand-drawn logo.
Two reflexes anyway, and they apply to a human-drawn logo too. Check priority before registering, a model that has seen millions of logos can hand you something very close to an existing one. And have the final result vectorised: a trademark office and a printer expect a clean file, not a 1024 pixel PNG.
What I actually do on a client production
A contract that describes the method, a proofread of the backgrounds, no artist names in the prompts and keeping the project: four habits that are enough in real life.
First habit: I tell my clients that I work with AI, and I write it in the quote. The subject becomes an argument about method instead of a secret that blows up six months later. And many commission contracts now contain a clause on the topic: you may as well read it before signing.
Second and third: no living artist's name in prompts, ever, and a dedicated proofread of the backgrounds before delivery. Those two remove most of the real risk at a cost of zero euros.
Fourth: I keep the project's full canvas, with the iterations. It serves protection, it serves when the client comes back nine months later asking for a variation, and it serves when you have to prove the volume of work done. On how to price that work, I detailed my grid in invoicing AI creative work to a client.
What to remember before your next generation
Use commercially without fear, protect yourself by adding documented human work, and watch four points: named artist styles, trademarks, real faces and the model's licence.
The summary fits in three lines. Your generations belong to you and you can sell them. Their protection against copying is weak as long as they stay raw outputs, and it rises with the amount of human work you put in. The real risks are elsewhere: in what the image shows, not in how it was made.
If you are starting out and the subject was holding you back, it should not any more. Start with your first image at 1 credit with Flux Schnell, look at what comes out, and keep those four watch points in a corner of your mind when you deliver. The exact price of each model is visible on the button before you even click, and the full detail is on the pricing page.
One last honest word: the law is moving on this subject, in Europe as in the United States, and a position held today may be refined tomorrow. What will not move is that documented human work puts you on the right side. It is also what makes the difference between a pretty image and an image that actually serves your project.