Who Owns AI Generated Images? ChatGPT vs Midjourney vs Gemini vs Canva vs Krater
Every platform says the image is yours. The next clause matters: who publishes it, who trains on it, who keeps a licence. Six platforms compared.
You own AI generated images on every major platform, as far as the platform is concerned: ChatGPT, Midjourney (paid plans), Google Gemini, Canva, Adobe Firefly and Krater all say the output is yours. What differs is what they keep and what the law gives you. Midjourney keeps a perpetual licence to your images and makes them public by default. ChatGPT, Gemini and Canva train on consumer content unless you opt out. Krater claims no licence and does not train on your work. And under US law a purely AI generated image has no copyright at all, so "owning" it means holding a licence to use it, not a right to stop others.
Key Takeaways
- "Who owns it" has two answers. The platform's answer is in its terms. The law's answer, in the US, is that nobody owns a purely AI generated image, because copyright needs a human author.
- OpenAI has the clearest wording: you own the output and OpenAI assigns you all its rights in it. Canva, Midjourney and Adobe say you own the output too. Google says "your content remains yours". Krater says you retain ownership of what you create.
- Ownership is not the whole story. Midjourney takes a perpetual, irrevocable, sublicensable licence to your prompts and images and publishes them by default. You own it, Midjourney can use it forever.
- Training is the quiet clause. ChatGPT consumer plans, Gemini consumer plans and Canva Free and Pro use your content to improve their models unless you switch it off. Krater and Adobe Firefly do not train on customer content.
- Copyright protects your human contribution: edits, selection, composition, added elements. The more of that you do, the more you actually own.
- Five questions cover it: do I own it, can I sell it, is it private, do they train on it, and what licence do they keep. The table below answers all five for six platforms.
Two meanings of "own"
When a platform says you own the output, it means that as between you and the platform, the platform is not claiming it. That is a contract. It settles who can use the image without getting sued by the platform.
Copyright is different. It is a right against everyone, and in the United States it only exists for works with a human author. The US Copyright Office's January 2025 report concluded that "prompts do not alone provide sufficient control" to make the prompter the author and that copyright "does not extend to purely AI-generated material". The D.C. Circuit in Thaler v. Perlmutter (March 2025) held that the Copyright Act "requires all eligible work to be authored in the first instance by a human being".
So on every platform in this article, the honest position is: the platform lets you own and use the image, and the law protects the parts you made yourself. OpenAI's own wording shows the gap. It assigns you "all our right, title, and interest, if any, in and to Output". The words "if any" are there because there may be no rights to assign.
Sources: US Copyright Office, Copyright and Artificial Intelligence, Part 2; Thaler v. Perlmutter, D.C. Cir. 2025.
Who owns AI generated images, platform by platform
| Platform | Do I own it? | Can I sell it? | Is it private? | Do they train on it? | What licence do they keep? |
|---|---|---|---|---|---|
| ChatGPT (OpenAI) | Yes. "You own the Output. We hereby assign to you all our right, title, and interest, if any" | Yes, all plans | Yes | Consumer: yes, unless you opt out in Data Controls. Business and API: no | None over output |
| Midjourney | Paid: "You own all Assets You create". Free: Creative Commons Noncommercial licence only | Paid: yes. Over $1M revenue needs Pro or Mega | No. "By default, Your Content is publicly viewable and remixable" | Broad licence covers reuse; no separate training opt out documented | Perpetual, irrevocable, sublicensable licence over prompts and images, survives termination |
| Google Gemini (Imagen, Veo) | "Your content remains yours." No explicit output assignment in consumer terms | Yes | Yes | Consumer: yes, with human review, unless Gemini Apps Activity is off. Workspace: no | Worldwide licence to host, reproduce and improve services |
| Canva (Magic Media) | "You own your Output", except output using Canva library content or AI audio | Yes | Yes | Free and Pro: on by default, opt out in Privacy Settings. Teams and Enterprise: never | Service licence; library content stays licensed |
| Adobe Firefly | "Customer owns Output" (licensing terms) | Yes | Yes, unless submitted to the public gallery | No. Models trained on licensed stock and public domain content | Service licence; gallery submissions get a wider licence |
| Krater | "You retain ownership of content you create using our service" | Yes | Yes. No gallery, no remixing | No. "We do not use your data to train AI models" | None claimed over outputs or prompts |
Sources: OpenAI Terms of Use, OpenAI Business Terms, Midjourney Terms of Service, Google Terms of Service, Gemini Apps Privacy Hub, Canva AI Product Terms, Adobe Generative AI User Guidelines, Krater Terms and Privacy.
ChatGPT: strong ownership wording, training on by default
OpenAI's terms are the most explicit of the six on ownership. You own the output and OpenAI assigns you whatever rights it has. There is no revenue threshold and no separate commercial tier. Two things to know: on consumer plans "if you do not want us to use your Content to train our models, you can opt out" in settings, which means the default is that your prompts and images are used; and images from ChatGPT carry both C2PA metadata and a SynthID watermark, so they are identifiable as AI generated even after editing. Business and API customers are excluded from training by default.
Midjourney: you own it, and so does everyone else
Midjourney says "You own all Assets You create with the Services to the fullest extent possible under applicable law", and that ownership survives cancelling. Then three clauses take most of it back for practical purposes. First, "by default, Your Content is publicly viewable and remixable", and anything made in a shared Discord space "is viewable by anyone". Stealth mode is only on Pro and Mega plans and comes with a "best efforts" promise, not a guarantee. Second, you grant Midjourney a perpetual, worldwide, irrevocable, sublicensable licence to your prompts and images that "survives termination". Third, free users do not get ownership at all, only a Creative Commons Noncommercial 4.0 Attribution licence, and businesses grossing over $1 million a year must be on Pro or Mega to use images commercially.
Google Gemini: yours, with human review unless you turn it off
Google's consumer terms do not contain an "you own the output" sentence. They say "your content remains yours" and grant Google a licence to host, reproduce and modify it to operate and improve its services. The Gemini Apps Privacy Hub is where the important part lives: "human reviewers (including trained reviewers from our service providers) review some of the data we collect", and that data is used to improve Google's generative models. Turning off Gemini Apps Activity stops your conversations being used this way. Images and video from Imagen and Veo carry an invisible SynthID watermark. Google Workspace accounts are excluded from training.
Canva: you own it, unless it touches the library
Canva's AI Product Terms say "you own your Output", with two exceptions: output that "modifies or incorporates Licensed Content" from Canva's asset library stays under Canva's content licence, and AI generated audio has separate resale limits. Free and Pro accounts are opted in to "improve our AI services for everyone" by default, with a switch in Privacy Settings. "Content from Canva Teams, Business, Enterprise, and Education users is not used to improve AI-powered features, and this cannot be turned on." Canva attaches C2PA provenance tags and bans removing them.
Adobe Firefly: owned, private, not trained on
Adobe's licensing terms state that "Customer owns Output". Firefly is marketed as commercially safe because its models are trained on licensed Adobe Stock and public domain content rather than on customer work. Outputs are private unless you submit them to the public Firefly gallery, which grants Adobe and other users a wider licence. Adobe attaches Content Credentials to generated images and you "must not remove, alter, or disable" them. One caveat for free users: Adobe's earlier generative AI terms granted Adobe a perpetual licence over free users' inputs and outputs "for any purpose"; check the current version of that page before relying on free tier output.
Krater: nothing claimed, nothing trained, nothing published
Krater's terms are short on this point on purpose. "You retain ownership of content you create using our service." "We do not use your data to train AI models. Your conversations, prompts, and generated content are used solely to provide you with our services." There is no public gallery, no remix feature and no licence grant over your prompts or outputs beyond what is needed to show them back to you.
One honest note. Krater runs image and video models from several providers in one workspace, so our terms add that "AI-generated content may be subject to the terms of underlying model providers." That is the same position as any platform routing to third party models; we would rather say it than hide it. What Krater itself does is simple: it takes no rights in your work and does not use it for training.
What you actually own: the human part
Because the law protects human authorship, the way to own more of an AI image is to contribute more to it. The Copyright Office lists what counts: perceptible human authored elements, and the creative selection, arrangement and modification of AI output. In practice:
- Generate several versions and choose. Selection is a human act.
- Edit the result: crop, recolour, combine, retouch. Modification is protectable.
- Add your own elements: text, layout, drawn parts, a series concept.
- Keep the working history. Prompts, drafts and edits are your evidence of contribution.
Doing this on a platform where drafts are private and untrained means your unfinished work is not feeding someone else's model or someone else's remix. On Krater you can create with any image model in /image or video model in /video, iterate in the same chat, and keep every version in Keep.
Frequently Asked Questions
Do you own images you make with ChatGPT?
Yes. OpenAI's terms say you own the output and assign OpenAI's rights in it to you. Consumer accounts allow OpenAI to train on your content unless you opt out, and the images carry C2PA and SynthID marks identifying them as AI generated.
Who owns Midjourney images?
Paid subscribers own their images. Midjourney also keeps a perpetual, irrevocable licence to them and to your prompts, and makes them public and remixable by default unless you are on Stealth mode. Free users only get a non-commercial Creative Commons licence.
Are AI generated images copyrighted?
In the US, not on their own. The Copyright Office and the D.C. Circuit both require a human author. Your edits, selection and additions to an AI image can be protected; the raw generation cannot.
Which AI image generator lets you keep the most rights?
Every paid platform here lets you own and sell the output. The differences are in what they take back: Midjourney keeps a perpetual licence and publishes by default; ChatGPT, Gemini and Canva consumer plans train on your content unless you opt out. Krater and Adobe Firefly claim no licence beyond running the service and do not train on customer content.
Can a company use AI images commercially?
Yes on all six platforms. The only revenue condition is Midjourney's: businesses over $1 million a year need a Pro or Mega plan. For commercial work, prefer a plan or platform that does not train on your inputs, since prompts and drafts often contain client material.
Bottom Line
Every major generator will tell you the image is yours. Read the next clause. Midjourney publishes and licenses your work; ChatGPT, Gemini and Canva train on it by default; Krater and Adobe do neither. And whatever the platform says, copyright in the US only covers what a human added. Own more by making more of it yourself, on a platform that keeps your work private.