Disney and Universal sue Midjourney for copyright infringement

In a spectacular development in the AI wars, Disney and several studios have sued the AI image-generating company Midjourney for copyright infringement (complaint here). To say that this has generated quite a stir would be a monumental understatement akin to declaring that Thunderbolts* was a bit of a flop. Not only have the usual generative AI and copyright nerds been interested, but judging by the media response, it would be fair to say that this has made it all the way to the mainstream media. It’s Disney, after all.

I think that the level of response has been caused by many reasons. The obvious one is the size and power of Disney, which is as big as you can get in media and entertainment, but there has also been a bit of surprise at two facts of the case: the timing and the defendant. There have been over 40 cases so far against different generative AI providers and developers, and so far the largest litigant was arguably the music companies suing Suno and Udio, so the absence of large media companies in the list of claimants had already been remarked upon repeatedly. Then there is Midjourney as the recipient. Whilst they have been party to one of the first cases filed by artists, most of the litigation since has taken place against tech giants, particularly Meta, OpenAI, Google, and Microsoft. Midjourney are relatively quite small when it comes to generative AI, and whilst they have a loyal user base, they cannot be compared with any of the other image providers.

I have to admit that I had sort of expected that the large media companies would not get involved in litigation for reasons that I have explained before, and although I did not fully rule it out, I thought that it would be unlikely for various strategic reasons. I still think that those hold, but the nature of the Midjourney lawsuit hint at what media companies like Disney will be expecting out of generative AI in the future, but more of that later.

Before starting, I’d also like to admit that I’ve been a paying customer of Midjourney for years. I’ve been using the platform since it was strictly run using Discord commands; to me it’s my favourite image model, which I use repeatedly for this blog and for teaching, and I am not afraid to admit that if they were to go out of business as a result of this lawsuit, I will be very sorry to see them go. So please read the following lines with that in mind. I’m also a Disney+ subscriber, but I’m less attached to that service. I loved Andor, but I’m still trying to erase The Acolyte from my brain.

The lawsuit

The lawsuit has been brought by Disney, Marvel, LucasFilm, 20th Century Fox, Universal, and DreamWorks against Midjourney for direct copyright infringement and secondary copyright infringement. No phoney DMCA claims here, no large number of unrelated or semi-related claims. This is direct and to the point; you can tell the claimants mean business. The argument is quite simple: Midjourney trained their model using images belonging to the claimants, and they are able to directly reproduce, publicly display, and distribute reproductions and derivative works of content owned by the distributors. Midjourney provides a service that allows users to infringe copyright by reproducing their characters on demand.

The complaint is filled with examples of prompts showing page after page of infringing outputs, from Darth Vader to the Minions. The complaint also has several examples of users sharing those outputs on social media, particularly the unofficial Midjourney subreddit. They also show several examples of search results that are able to display outputs generated by its users of several characters owned by the claimants.

This is quite possibly the most relentless and complete complaint in the AI Wars. After all, it’s Disney, they don’t come unprepared. I could have a long analysis of the case, I’m sure that there will be time for that in the coming months, but on first reading this is exactly what one would expect from a battle-hardened legal team. I would have liked to give Midjourney some hope, but barring a spectacular out of court settlement, it’s not looking good for the defendants. As I told New Scientist (and they actually published it!), “It’s Disney, so Midjoureny are fucked, pardon my French”. Quoting Lio Partagaz from Andor, “Basically, it’s bad luck Midjourney”.

Why am I so sure of a Disney victory? Well, when it comes to copyright lawsuits, size does matter, the large media companies involved can outlawyer a small AI company, it’s a matter of quality of legal advice and resources, it’s like a cheat code in which you get unlimited lawyers. And betting against the Mouse is always a bad idea, regardless of the merits of the case…

Needless to say, the case looks very strong. For a long time we have known that image generators were able to produce potentially infringing outputs, unlike LLMs, these have been substantial enough for over a year now, and some of these could even pass the realm not only of infringing derivatives, but often direct reproductions of works. This is why this case is outright different from all of the others, with images you can produce infringing outputs, these are smoking guns. The difference with the other cases is that it will be much more difficult for Midjourney to claim fair use, because that argument applies mostly to the inputs, not so much on the outputs. Some of the outputs may end up being transformative, but a lot of outputs are pretty much in outright infringing territory.

There are a couple of things that may leave an opening for arguments by the defendant. Firstly, Disney seems to be going with the argument that the Midjourney model “stores” images from the claimants. This is not technically correct, but the assumption here is that Midjourney operates more like a direct infringing database that can reproduce specific works on demand. In some ways this is correct with regards to characters, the model knows what Elsa and Baby Yoda look like, and can generate outputs that resemble the inputs, and one could definitely argue that these are infringing reproductions, close enough to meet the substantial similarity requirements. But Midjourney does not store those images and serves them, image models know what a stormtrooper looks like and can reproduce it. Is it storage? I don’t think so. That may end up not having an effect overall, but it may be an opening for the defendants.

Secondly, the claimants are leaning heavily on the argument that Midjourney is specifically trained to produce infringing outputs, and that it profits from that fact, alleging that this is a draw to consumers. They argue that sometimes works they own show up in the “Explore” page, and that together with the search tool which displays character content on demand, it’s evidence that the defendants have knowledge that their service is being used for this purpose. I think that here they’re setting up a Grokster inducement argument for secondary infringement.

This could actually prove a good avenue to explore. As I mentioned, I’ve been an habitual user of Midjourney for over 2 years, and what is described in the complaint doesn’t track with my own experience. Sure, from time to time you see a character in the “Explore” pages, but they are usually very stylised and original, more into transformative territory if you ask me. And these tend to be very rare, at least in my experience. From a personal perspective I’ve produced thousands of images, and the only time I have used the system to produce characters have been precisely for teaching, writing, and presenting. So I think that the strongest defence that Midjourney will have is that it is a service that has substantial non-infringing uses, and it will be interesting if they can back this up with statistics.

But they’re against Disney…

Why Midjourney?

A question I have been asked repeatedly during the last couple of days is that people are baffled at why Disney is attacking Midjourney, and why now. I don’t know is the real answer, but I have a theory. Disney knows that AI is here to stay, they state it in the press releases accompanying the news. One of their representatives told CNN:

“We are bullish on the promise of AI technology and optimistic about how it can be used responsibly as a tool to further human creativity,” Horacio Gutierrez, Disney’s senior executive vice president and chief legal and compliance officer said in a statement to CNN. “But piracy is piracy, and the fact that it’s done by an AI company does not make it any less infringing.”

Anyone who thinks that this lawsuit will kill generative AI is sorely mistaken, Disney does not want to destroy AI, it wants to use it. They’re like Isildur saying no to Elrond, they’re like Boromir saying “let us use the weapon of the enemy”. Wait a second, why am I referencing Lord of the Rings? Wrong franchise, but I digress…

So my theory is that this is sending a message to all of the other AI image generators and AI developers to be careful, stay in line, mind your output filters, or you could be next. Midjourney is small, it is expendable in the grand scheme of things. Disney may want to work with OpenAI, Google, Microsoft, and Meta in the future, so no need to start attacking potential partners, particularly as they have made very clear that they will be using generative AI themselves. What I think will happen is that they will be using their extensive catalogue to train their own models, or perhaps to provide licensed and official fine-tunes. And the goal will be to fire a bunch of creatives and produce new content based on the works they already own.

So Midjourney could end up being like the Napster of the AI Wars, the first fatality.

As to why now? The models were able to produce potentially infringing outputs for over a year, maybe even more than that, they must have been ready to go for a while now. I think that the rise of video models may have really put the fear of God into some execs, so it may seem like the right time to act. Disney are ready to play ball, come and join us, or else.

Concluding

These are exciting times to be an AI Copyright researcher, but they’re also tough. It’s becoming impossible to keep track of all of the cases, particularly if you don’t have an army of research assistants and very limited resources. But this is a big one, and definitely it is one to follow closely, even though it may just be a chronicle of a death foretold, there is just going to be too much of a disparity in resources.

But then again, you never know. Perhaps there’s a small hidden fault in the Complaint. Perhaps this fault will unleash a catastrophic failure, perhaps it’s like a fault in the exhaust pipe of the Death Star. Perhaps it is small enough that a plucky underdog in a single fighter can exploit with a one-in-a-million shot. Nah, that only happens in movies.

Anyway, begun, the AI Wars have. We’re in the endgame now.

I can do this all day…