llama toast

A llama wearing a beret eating avocado on toast at a Parisian cafe, in the style of Van Gogh.

Artificial intelligence generated art is here to stay. In recent weeks OpenAI has given wider access to its AI graphics tool called DALL·E, and that has been followed by an explosion of imaginative, wacky, baffling, bizarre, and sometimes beautiful art being shared with the wider public. The success of the tool has been such that OpenAI has announced that it will be commercialising its image generation platform, so you if you want to create some AI art you will be able to purchase credits that can be used to make prompts to generate art, beta testers have been given 100 credits as a token of appreciation.

This announcement is very interesting for many reasons. For starters, it is becoming clear that we are starting to witness some of the business models that are being implemented using AI, some of the companies involved will be trying to recover some of their investment by selling access to the tools they have developed. This is also perhaps a preview of the future of creativity, where humans will be relying more and more on AI in the creative process. We may be about to see a new type of creative worker, the AI whisperer, a person who has a talent of getting the AI to produce a worthy piece of work.

But from my perspective one of the most interesting elements, and one that I have been writing for many years now, and it is the copyright status of all of these creations. Who owns these DALL·E outputs? Is it OpenAI? The person who writes the prompts? Or is it nobody at all?

The first place to go to answer these questions is DALL·E’s own terms of use. One thing you need to know first is that when using the tool, you have two options, you can upload your own image to have it modified by the AI, or you can write a prompt to generate your own image. These are referred to in the terms of use as Uploads, Prompts, and Generations respectively.  OpenAI allows for commercial use of the Generations here:

Use of Images. Subject to your compliance with these terms and our Content Policy, you may use Generations for any legal purpose, including for commercial use. This means you may sell your rights to the Generations you create, incorporate them into works such as books, websites, and presentations, and otherwise commercialize them.”

So far so good, you can use the Generations you have created in other media (such as using my Generations in this blogpost). But doesn’t that imply that they own the Generations? If they’re allowing me to make commercial use of the generated content, it means they own it, right? This appears to be what they’re claiming. Here’s the paragraph relating to ownership of the generated images:

Ownership of Generations. To the extent allowed by law and as between you and OpenAI, you own your Prompts and Uploads, and you agree that OpenAI owns all Generations (including Generations with Uploads but not the Uploads themselves), and you hereby make any necessary assignments for this. OpenAI grants you the exclusive rights to reproduce and display such Generations and will not resell Generations that you have created, or assert any copyright in such Generations against you or your end users, all provided that you comply with these terms and our Content Policy. If you violate our terms or Content Policy, you will lose rights to use Generations, but we will provide you written notice and a reasonable opportunity to fix your violation, unless it was clearly illegal or abusive. You understand and acknowledge that similar or identical Generations may be created by other people using their own Prompts, and your rights are only to the specific Generation that you have created.”

That is some really interesting wording, and I find this paragraph to be a work of art. Firstly, you own your own Uploads and Prompts to the extent allowed by law. This is because a prompt may be too short to have any sort of protection. So let’s say I just type “Cyberpunk llamas”, I don’t think that prompt is worthy of protection, but “a llama wearing a spacesuit in a futuristic city scape with flying cars flying above, digital art”, is detailed enough that it just might have protection in its own right (more on that later). By the way, here is the image:

But what about the generated image itself? Here is the interesting part, OpenAI owns the image, but you will notice that they do not claim any copyright ownership over the image, just that they own it by means of contract. The wording is clever, it says “you agree that OpenAI owns all Generations”. So by using the service, you agree to the terms of use, and the terms of use state that you agree that OpenAI owns the images, regardless of whether those images have copyright or not. OpenAI grant you the permission to reproduce and display Generations (as I’m doing right here). They also state that they will not make any copyright claims against the user, which is nice as there may not be a copyright claim to begin with.

The final sentence in that section is quite interesting as it recognises the possibility of people coming up with similar generated images based on similar prompts. This is an elegant way of bypassing independent creation, but also of tackling the idea/expression dichotomy.

So are there any copyright issues? At the moment OpenAI seems to cleverly bypass most copyright questions through contract, making only oblique references to ownership in an IP sense. This is a large conglomerate with extremely capable lawyers, so I’m certain they must be aware that these images have a tricky copyright path ahead of them. For the most part, the legal consensus appears to be that the images do not have any copyright whatsoever, and that they’re all in the public domain. If that was the case, then we are now presented with an endless trove of public domain images featuring llamas, robots, and heavy metal cats. This is precisely why OpenAI is asserting its ownership of the images based on contract law, and not on copyright. The question may come at some point as to whether they can impose this obligation regarding ownership of an image that is in the public domain.

The situation may be different in the UK, where copyright law allows copyright on a computer-generated work, the author of which is the person who made the arrangements necessary for the work to be created. This, in my opinion, is the user, as we come up with the prompt and initiate the creation of the specific work. I think that there may be a good case to be made that I own the images I create in the UK. In this case, what is happening with DALL·E’s terms and conditions is that I am agreeing to transfer that copyright to them, this is why they specify that “you hereby make any necessary assignments” regarding their ownership of the images. Clever!

The same thing would apply to EU copyright law, where copyright exists on any original work, and the work is original if it is an intellectual creation. I would argue that this is true for some more complex prompts, for example, “a llama playing poker in a blue room, with a painting on the wall, and a window with the sun shining in” [picture here]. But I would also argue that a more basic prompt would not be an intellectual creation, and therefore not original, and thus not having copyright. Take for example “a llama recording a podcast“.

Concluding

The commercialisation of AI works is here to stay. I have started noticing a marked increase in AI-generated artworks popping up all over NFT marketplaces, some of it I will admit, is quite beautiful. For example, I just bought a very stunning AI work for about $3 USD (2 TEZ), I am perfectly aware that this work may not have copyright, but I thought it was quite pretty, so why not? I even minted one of my AI llamas here, all in the name of research (it can be yours for 2 TEZ).

All of this commercial action will inevitably lead to litigation, perhaps a commercial company will try to challenge OpenAI’s property claims, or perhaps OpenAI itself will sue. On the meantime, I leave you with a picture of a llama in the metaverse.

 


17 Comments

Avatar

andrewducker · July 26, 2022 at 3:32 pm

Is there anything to be concerned here with the ownership of the inputs that went into creating the DALL-E system in the first place? It’s unlikely to be as blatant as with Github replicating people’s code verbatim, but if it produces someone’s pre-existing art then who is responsible, if anyone?

    Avatar

    Bex · July 27, 2022 at 11:29 am

    I was pretty sure that that was the biggest copyright question as well. You can declare that Dall-E’s works are in the public domain all you like, but OpenAI is still making a profit from them, and if it’s trained on imagery scraped from the internet… that’s hella shaky legal ground.

      Avatar

      matchateanice · July 30, 2022 at 11:19 am

      but is it working with images scraped from the net or creating them all from scratch? How can we know for sure? I was told that it was the latter. If not, that is indeed hella shaky legal ground and very worrying.

        Avatar

        andrewducker · July 30, 2022 at 6:49 pm

        It’s trained on images scraped from the internet, and uses that training to then create them from scratch.

    Avatar

    Andres Guadamuz · July 28, 2022 at 9:28 am

    That’s a great question, stay tuned, writing an article about this.

Avatar

Henr · July 27, 2022 at 9:35 pm

Intellectual Property laws should be renewed to deal with these datamining-based machines. Leaving the question of mass-scraped uncredited works aside, If OpenAI “owns” any image they reproduce but do not enforce clear copyright over it, while also allowing commercial use for NFTs, blogs, books and such, I could make NFTs based on existing fictional/real characters and sell them just like it charged me to reproduce their images.

Also the UK laws open up to some horrible long term consequences. If I had money to make an AI with enough computing power to generate pictures really fast through randomised prompts, eventually I would own copyright to every previously-unused combination of pixels possible in a 2k/4k/8k… canvas. You’re signing up to an era of digital hell.

Avatar

Andy J · July 28, 2022 at 7:02 pm

I’m not so sure that the contract approach really is that smart. The problem with a contract is that it only binds the actual parties to the contract. So if a third party (say a reader of this blog) takes you image of a llama in the metaverse and exploits it economically, neither you (the author and owner of the prompts) nor Open AI have any cause of action against that reader.

And I would also suggest that the situation in the UK is not quite as benevolent towards OpenAI as you suggest. Yes, section 9(3) makes the prompter the first owner of copyright in the generated image, but OpenAi’s contract terms cannot be deemed to assign ownership to them because section 90(3) CDPA specifically requires that this is done in writing and signed by the first owner, and a click-through agreement alone isn’t going to meet that criterion.

Interesting issue, though. And far too early to talk about changes to the law, in my view. We need a few more test cases (like the DABUS litigation) to see the extent of the problem.

    Avatar

    Andres Guadamuz · July 29, 2022 at 11:13 am

    Excellent points. I’d probably argue that OpenAI could intervene with third party rights, but that could be very jurisdiction dependent.

    As to 90(3), absolutely true, but the terms and conditions place the onus on you to make that assignment, it is cleverly worded. Will it stand up in court? There’s only way to find out.

Avatar

Kenee · July 30, 2022 at 12:18 am

Are DALL-E users legally allowed to modify images generated by DALL-E?

    Avatar

    Andres Guadamuz · July 30, 2022 at 10:54 am

    The terms do not specify, but as you have permission to reuse the images, I think that you should be able to.

Avatar

matchateanice · July 30, 2022 at 11:16 am

Great article. Am I able to use the DALLE images for print on demand? I’m sure that a lot of people will be wondering the same thing…

    Avatar

    Andres Guadamuz · July 31, 2022 at 10:21 am

    The images you have generated? Yes. For other people’s generations you need permission from OpenAI.

      Avatar

      matchateanice · August 10, 2022 at 4:02 pm

      Thanks so much for the reply. The last post makes a really good point. How does someone apply the appropriate credit for these images? I’m not even sure that, for example, on RedBubble it would be possible to credit as Open AI / your brand name. Knowing how to credit would help many print on demand creators like myself. Any advice would be greatly appreciated. 🙂

Avatar

forbiddenera · July 31, 2022 at 5:53 pm

‘the person who made the arrangements necessary for the work to be created’ to me also sounds like the software developer(s) so potentially could maybe be argued that both the author of the prompt and the developers might have rights. But then again, Photoshop doesn’t (and probably couldn’t, assuming it was their goal) assert any rights over anything created with their software by someone.

DALL-E is a program; it may have been trained on previous inputs but that simply produced differential equations akin to those used in say, a blur effect in Photoshop. The inputs it requires to create an output, the prompt, to me, would be akin to any inputs Photoshop would require to create an output. If you treat it like any other computer program that takes your inputs to create outputs, which is exactly what it is, the outputs should be 100% entirely your creation and copyright, whether ‘generated’ or not. Plenty of computer programs ‘generate’ some sort of output with various forms of input – just because the equations here are magnitudes of order larger and more complex, IMHO, does not change what it is.

Definitely an interesting topic. I feel like, perhaps that aside, with their contract saying basically you can do whatever and we will never assert a (C) claim could potentially void any ability to make such a claim, if not by written law perhaps by precedent set by a case.

Avatar

Matt · August 7, 2022 at 10:29 am

Just want to double check. If I got this right, images created from a prompt can be sold on POD sites as prints, t-shirts etc. ? So if that’s the case who should be stated as creator ? Is it legal to list the product under your name ?

It also says: “You must clearly indicate that images are AI-generated – or which portions of them are – by attributing to OpenAI when sharing, whether in public or private.” How do one do that and is that mandatory ?

Also what if someone just takes a screenshot and starts selling your generated images, you can probably not do anything about that or I’m missing something ?

Thanks in advance

    Avatar

    matchateanice · August 10, 2022 at 4:03 pm

    Yeah, I would really like to know what others think about these questions too. Would be brilliant to have some answers. Greatly appreciated in advance.

    Avatar

    Andres Guadamuz · August 10, 2022 at 9:05 pm

    Sorry, had to delete the previous answer as it pertained to the API itself.

    • “Is it legal to list the product under your name ?”

    You don’t own the images according to their own terms, so while you can say that you wrote the prompts (and you may own them), the image itself is not yours, you get a licence.

    • “You must clearly indicate that images are AI-generated – or which portions of them are – by attributing to OpenAI when sharing, whether in public or private.” How do one do that and is that mandatory ?”

    It’s in their terms, you have to comply with the terms to use the service.

    “Also what if someone just takes a screenshot and starts selling your generated images, you can probably not do anything about that or I’m missing something ?”

    You can tell OpenAI that someone is using their image, but remember, the images may not even have copyright to begin with.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: