This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions.

We read the Midjourney Terms of Service so you don’t have to. In a “blink and you’ll miss it” clause, Midjourney can and will sue you for copyright infringement in at least one scenario.

Midjourney has rapidly evolved to become for AI-generated artwork what ChatGPT has been for AI-generated text (at least for now, until OpenAI releases its imminent “multimodal” update which will enable users to interact with images, videos, and even voice).

Much has been debated about the nature of copyright as pertains to AI training. AI tools like ChatGPT, Google Bard, Anthropic’s Claude, Inflection’s Hey Pi fundamentally operate the same: they trawl the internet and use that data to train their Large Language Models (LLMs).

How AI Large Language Models (LLMs) work

To clarify, the generated output does not use the source training material; rather, the output is novel, even if there is a non-zero probability that it may not be unique:

LLMs operate based on statistical patterns in the training data, predicting sequences of words based on the context provided; in this sense, these AI tools are context-aware, even if the context is derived from probability and not true "meaning.” The result is that ChatGPT, for instance, is more likely to output a sentence claiming that “the cat sits on the floor” rather than that “the cat sits on the hamburger.” Art tools like Midjourney do similar training, albeit for image generation. Hence if you prompt (with written words) Midjourney to create artwork “in the style of Monet,” for instance, it will indeed generate precisely such a piece.

Although the output is unique at least in the sense that it does not directly regurgitate any of the source training data, it is probabilistically possible that output can coincidentally closely mirror, if not precisely replicate, some of that training data, if only by mere chance.

Can LLM output infringe on copyright?

This is a much bigger discussion for another article, but a significant legal debate has emerged both for AI-generated written output as well as AI-generaed image output: if the training data for these AI platforms’ LLMs utilized copyrighted material in the first place, then is it possible for someone — an artist, say — to sue either (A) the AI tool itself and/or (B) the end user of that tool if the artist claims the tool was used to infringe on the artist’s copyright?

Under US Copyright Law, intent is not a required element to prevail under a copyright infringement claim. In other words, somebody does not need to intend, or knowingly infringe upon, somebody’s work to be liable for copyright infringement.

This means that an artist — writer, musician, painter, etc. — can always sue somebody for copyright infringement, and proving intent or knowledge of that infringement is not required to prevail. And of course, to the surprise of nobody, the artist can indeed sue the AI tool itself, like Midjourney, for instance.

Without delving into the unsettled law on whether a plaintiff in such a lawsuit would prevail, there’s still the question of what happens if somebody uses Midjourney to intentionally infringe on somebody else’s work?

Midjourney’s Limitation of Liability and Indemnity clause

Paragraph 10 of Midjourney’s Terms of Service — “Limitation of Liability and Indemnity” — outlines the legal ramifications of knowingly engaging in copyright infringement on the platform. It states, in relevant part, with emphasis added:

"If You knowingly infringe someone else’s intellectual property, and that costs us money, we’re going to come find You and collect that money from You. We might also do other stuff, like try to get a court to make You pay our attorney’s fees. Don’t do it."

This is an extremely profound detail which is easily missed, first, because in general people never read such terms of use, and second, because even if you read it, and especially if you don’t know what you’re looking for, this could be easily glossed over.

That said, Midjourney has written their Terms of Service in fairly plain English, and it is remarkably unambiguous: Simply put, Midjourney is introducing a copyright infringement intent element: if you knowingly use Midjourney to create artwork that you know infringes on the artwork of another, and if that other person then sues Midjourney for copyright infringement, then Midjourney can indeed come after you for any monetary damage and attorneys’ fees suffered by Midjourney in that lawsuit.

This is profoundly important and cannot be reiterated strongly enough: Do not use Midjourney to knowingly infringe on another artist’s artwork.

Three different copyright infringement scenarios

It’s worth distinguishing the various scenarios of liability between US Copyright Law (no intent necessary for copyright infringement) and Midjourney’s Terms of Service (knowledge necessary); there are three such scenarios:

• (A) Knowingly infringe: If an end user knowingly employs Midjourney to create artwork that infringes upon another artist's work, the legal repercussions are multifold: (i) The infringed artist has the right to sue both the end user and Midjourney under US Federal Law for copyright infringement. Concurrently, Midjourney can initiate legal actions against the end user for knowingly infringing, as stipulated in its Terms of Service.

• (B) Unknowingly infringe: If the end user infringes on another artist’s work, but without knowledge or intent of doing so, then, the infringed artist still retains the right to sue both the end user and Midjourney under Federal law. However, per Midjourney’s Terms of Service, the end user is safe from Midjourney coming after the end user.

• (C) Unknowingly infringe but a reasonable person would have known: It’s worth mentioning a third scenario that might involve an end user of Midjourney unknowingly infringing, yet with prevailing evidence that a reasonable individual would have discerned the infringement. Simply put, if the end user didn’t knowingly use Midjourney to infringe on somebody else’s artwork, the legal question might hinge on whether the end user “should have known,” or whether a reasonable person would have known. The legal dynamics in this scenario could be intricate, potentially furnishing both Midjourney and the infringed artist with a basis for legal action against the end user.

Other platforms’ similar “User Generated Content” clauses

It's worth noting that clauses like this one from Midjourney's Terms of Service are not rare, but rather are a common legal safeguard employed by platforms hosting User Generated Content (UGC). These clauses act as a protective shield, minimizing platforms’ liability in the face of potential legal challenges stemming from content created by its users.

That said, the landscape becomes slightly nuanced when we venture into the realm of AI-driven platforms like Midjourney. Unlike traditional UGC platforms, AI-driven platforms have an added layer of complexity due to the machine learning models they employ, which could potentially generate content that flirts with the edges of copyright infringement. By emphasizing the knowledge or intent of infringement, Midjourney is not merely following a legal precedent set by other UGC platforms, but also navigating the intricate interplay between AI, creativity, and legal accountability, thus underlining the necessity for users to tread cautiously in their creative exploits on such innovative platforms.

While it is far too early, and the law far too unsettled, to definitively arrive at any conclusions, a common sense approach to AI-generated work, whether written, image, or otherwise, seems clear: do not knowingly infringe on somebody’s creation, or you may open yourself to liability from the AI platform you used in the first place.

To learn more about the intersection of AI and copyright law, don’t hesitate to contact us.