a day ago
Hollywood Sues AI Firm For Copying Characters With No License
Disney and Paramount's lawsuit, against AI image generator Midjourney, could reshape AI's IP future.
On Tuesday, Disney and Universal filed a joint copyright infringement suit against Midjourney, one of the most widely used generative AI image platforms. The complaint, filed in California, alleges that Midjourney knowingly trained its AI models on copyrighted film stills and character images – without permission. Noting that the resulting outputs that users produce are very similar to copyrighted content.
But what's at stake here isn't just retroactive damages or royalty reimbursements. According to Yelena Ambartsumian, an AI governance and IP lawyer and founder of Ambart Law PLLC, the real concern is forward-looking.
'I think Disney and Universal's major concern is not the licensing fees and royalties they have missed out on from Midjourney's training datasets,' Ambartsumian said in an interview, 'but rather Midjourney's soon-to-be-released commercial AI video service.'
In other words, this isn't just about Mickey Mouse showing up on a knock-off poster. It's about synthetic video fakes.
Midjourney's AI models currently generate static images, but the company has been promising a text-to-video upgrade that could generate full-motion scenes. This legal action confirms that the movie studios aren't waiting to see if those synthetic reels might mirror their IP.
As Ambartsumian notes, 'Their complaint for copyright infringement seeks an injunction, not only of Midjourney's image-generation service but also of the forthcoming video-generation service.'
The studios' complaint even includes side-by-side comparisons of model outputs and iconic animated film stills – some of which are virtually indistinguishable to the untrained eye.
This collection of AI generated images was included on page six of publicly disclosed legal filing.
Until now, generative AI companies have often leaned on the idea of 'transformative use' under fair use doctrine. That is, if an output changes the source material with some type of novel addition, it might not violate the copyright on the original. But this suit could punch a gaping hole in that argument.
'This lawsuit fundamentally breaks the 'transformative' defense,' said Dr. Manny Ahmed, founder and CEO of OpenOrigins, a firm focused on content authenticity. 'Most lawsuits have focused on copyright infringement from content scraping and AI model training, whereas this lawsuit targets the reduction of copyrighted material by a model.'
That's a critical distinction. Rather than arguing about whether the model scraped copyrighted images, Disney and Paramount are highlighting what it can now produce.
Based on the details listed in the filing, the movie creatives are more interested in what the AI models can produce once trained, and less interested in the copyrighted materials used for that training. This is a significant shift that could be a watermark moment – both figuratively and literally – regarding how the courts handle infringement cases going forward.
One of the murkiest aspects of IP law regarding AI involves 'style.' Examples of the concept are the unique stylistic renderings of characters on The Simpsons or the distinct look and style of movies from Studio Ghibli. The question is can an AI model mimic the artistic style of a movie without using exact characters? If so, is the generated content protected?
Dr. Ahmed points out that this case might broaden the legal conversation.
'What remains to be seen is whether the argument being made by Disney and NBCU will also apply to 'styles' and not just distinctly identifiable characters/IP assets,' he said. 'If it does, this would impose even broader restrictions on the outputs of these models.'
If that happens, generative AI firms may have to navigate a minefield of visual boundaries—not just around iconic characters like Elsa or Shrek, but around entire cinematic aesthetics.
Some in the AI space have expressed frustration that Midjourney didn't implement better safeguards to avoid this outcome.
'Midjourney may have a point—that it's difficult to know the provenance of the millions of images its diffusion models are trained on,' Ambartsumian told me. 'But that's not an excuse for failing to implement safeguards post-training, after learning that the models can create outputs that likely infringe on copyrighted works. Why Midjourney did not implement those safeguards, before deployment, is baffling to me.'
Ahmed agrees that technical and policy steps are available – and increasingly necessary.
'There are two paths Midjourney could pursue,' he said. 'One, Midjourney should consider collaborative relationships when it comes to training. Licensing content is a minimum standard. Two, when it comes to the outputs, Midjourney should agree up front with its licensors what kinds of outputs are allowed.'
That kind of upfront agreement could enable controlled replication in some cases – like promotional content or fan art – but only with explicit consent from the rights holder.
Even if Midjourney settles or wins this case, the broader AI industry will need to reckon with its implications. According to Ahmed, the suit underscores a growing appetite for clear boundaries – and legal consequences – when it comes to the reproduction of high-value IP.
'Having clarity around what uses are restricted would also give rights holders greater confidence in licensing their content for training,' he said.
That clarity may be coming through court orders rather than negotiation tables – for now. But either way, the message is clear: AI companies that overlook copyright concerns may find themselves cast in a very expensive sequel.
