Midjourney v6 has intensified a familiar AI copyright argument: when an image generator can recreate familiar characters, brands and movie-like scenes with simple prompts, who carries the legal risk?
After complaints about possible copyright infringement, Midjourney updated its Terms of Service with unusually direct language. The company now makes clear that users may be held responsible if they knowingly use the service to infringe someone else’s intellectual property.
What changed in Midjourney’s terms
The update, added on Dec 26, 2023, places more responsibility on the person generating the image. Midjourney’s position is that its service should not be used for copyright infringement, even if the tool makes that easy to attempt.
The Terms of Use state:
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 legal fees. Don’t do it.
The language matters because it reframes the risk around AI image generation. Midjourney is not only warning users against misuse; it is also saying that costs created by intentional infringement may be pushed back onto the user.
That does not settle the larger copyright debate. It does, however, show how seriously the company is treating claims that v6 can produce outputs close enough to copyrighted material to raise legal and commercial concerns.
Why v6 triggered fresh concern
The original issue centered on alpha testing of Midjourney v6. Users shared examples on X suggesting that the model could generate images very close to copyrighted original images, including recognizable brands, well-known people and scenes resembling movie material.
One prominent example involved Joaquin Phoenix in the 2019 Joker movie. A prompt using the wording "Joaquin Phoenix Joker movie, 2019, screens from movie, movie scene" reportedly returned an image that closely matched a movie scene, with the main differences appearing in lighting and color.
Other examples showed generated superhero-style images that did not always match a specific film scene, but could plausibly look like one. Some outputs were described as close to film originals, with changes such as camera angle or posture.
The concern is not only that these images are possible. It is that the prompts described in the source are simple: a movie name, perhaps a year, and terms such as "movie scene" or "screenshot from a movie". The article also describes similar experiments with Gandalf from "Lord of the Rings", where some results looked close to the original Gandalf.
Overfitting is the technical concern
The source points to a possible explanation: overtraining or overfitting. In plain terms, this can happen when a model is trained repeatedly and intensively on the same data, increasing the chance that the output resembles material from the training set.
That possibility is important because the examples are not framed as ordinary resemblance alone. The worry is that some generated images appear so close to originals that they may look like modified versions of training data.
The source also notes that studies have indicated this can happen, and that ChatGPT can show signs of text overfitting. For image tools such as Midjourney, the concern becomes especially visible when outputs resemble widely recognizable films, characters or branded material.
There is a different kind of risk with less famous images, such as stock photography. A user might create something that infringes copyright without realizing it because the source material is not widely known. Midjourney’s terms, as described in the article, refer specifically to intentional copyright violations.
Artists and rights holders see a market problem
Illustrator and film concept artist Reid Southen, who has worked for Marvel Productions, criticized Midjourney on X and made a video showing plagiarism experiments with v6. According to the source, he said images were deleted from his account because of his comments, and he also said his account was banned and he was removed from the Midjourney Discord group.
Southen accused Midjourney of "illegally using copyrighted IP without a license." He also said the AI software can create "exact copies of copyrighted IP, as well as infinite derivatives." His concern is that artists could end up competing against outputs derived from the same market their work helped create.
He also raised brand confusion as a risk, writing: "Or what about brand image issues and consumer confusion when 50% of all Marvel stuff online ends up AI knockoffs?"
The article also notes that simple prompts can produce results resembling classic artworks such as the Mona Lisa. In that case, the legal issue is different because the image is in the public domain due to its age.
The unresolved legal question
Midjourney is likely to argue that its image generator is a tool, comparable in some ways to a blank canvas that a user fills through prompts. A tool can be used for lawful or unlawful work, and the source compares this to Photoshop, which can also be used to make legitimate or illegitimate content.
The opposing concern is that Midjourney v6 can create images close to protected works because it was trained on images that Midjourney has not licensed. That makes the training process itself central to the dispute, not just the user’s prompt.
Much depends on whether courts decide that unauthorized use of data for AI training qualifies as "transformative use" and therefore "fair use." The source says numerous lawsuits are pending, and that Midjourney is already involved in at least one.
For now, the practical message is clear. Midjourney v6 may be capable of producing outputs that look uncomfortably close to protected material, and Midjourney’s updated terms tell users that intentional intellectual property infringement can come back to them directly.