Midjourney’s new AI-generated video tool will produce animated clips featuring copyrighted characters from Disney and Universal, WIRED has found—including video of the beloved Pixar character Wall-E holding a gun.
It’s been a busy month for Midjourney. This week, the generative AI startup released its sophisticated new video tool, V1, which lets users make short animated clips from images they generate or upload. The current version of Midjourney’s AI video tool requires an image as a starting point; generating videos using text-only prompts is not supported.
The release of V1 comes on the heels of a very different kind of announcement earlier in June: Hollywood behemoths Disney and Universal filed a blockbuster lawsuit against Midjourney, alleging that it violates copyright law by generating images with the studios’ intellectual property.
Midjourney did not immediately respond to requests for comment. Disney and Universal reiterated statements made by its executives about the lawsuit, including Disney’s legal head Horacio Gutierrez alleging that Midjourney’s output amounts to “piracy.”
It appears that Midjourney may have attempted to put up some video-specific guardrails for V1. In our testing, it blocked animations from prompts based on Frozen’s Elsa, Boss Baby, Goofy, and Mickey Mouse, although it would still generate images of these characters. When WIRED asked V1 to animate images of Elsa, an “AI moderator” blocked the prompt from generating videos. “Al Moderation is cautious with realistic videos, especially of people,” read the pop-up message.
These limitations, which appear to be guardrails, are incomplete. WIRED testing shows that V1 will generate animated clips of a wide variety of Universal and Disney characters, including Homer Simpson, Shrek, Minions, Deadpool, Star Wars’ C-3PO, and Darth Vader. For example, when asked for an image of Minions eating a banana, Midjourney generated four outputs with recognizable versions of the cute, yellow characters. Then, when WIRED clicked the “Animate” button on one of the outputs, Midjourney generated a follow-up video with the characters eating a banana—peel and all.
Although Midjourney seems to have blocked some Disney and Universal-related prompts for videos, WIRED could sometimes circumvent the potential guardrails during tests by using spelling variations or repeating the prompt. Midjourney also lets users provide a prompt to inform the animation; using that feature, WIRED was able to to generate clips of copyrighted characters behaving in adult ways, like Wall-E brandishing a firearm and Yoda smoking a joint.
The Disney and Universal lawsuit poses a major threat to Midjourney, which also faces additional legal challenges from visual artists who allege copyright infringement as well. Although it focused largely on providing examples from Midjourney’s image-generation tools, the complaint alleges that video would “only enhance Midjourney ability to distribute infringing copies, reproductions, and derivatives of Plaintiffs’ Copyrighted Works.”
The complaint includes dozens of alleged Midjourney images showing Universal and Disney characters. The set was initially produced as part of a report on Midjourney’s so-called “visual plagiarism problem” from AI critic and cognitive scientist Gary Marcus and visual artist Reid Southen.
“Reid and I pointed out this problem 18 months ago, and there’s been very little progress and very little change,” says Marcus. “We still have the same situation of unlicensed materials being used, and guardrails that work a little bit, but not very well. For all the talk about exponential progress in AI, what we’re getting is better graphics, not a fundamental principle solution to this problem.”