The Bottomless Pit of Plagiarism

Last week, Disney and NBC Universal marched into a Los Angeles courthouse with a 140+page complaint. Their target? Midjourney, an AI company that turns text into images.

The lawsuit, some have said, reads like a pop culture roll call: Darth Vader, Elsa, Shrek, and 196 other beloved characters all make appearances.

Disney's claim is pretty simple: Midjourney turned their creations into what they call "a bottomless pit of plagiarism."

The price tag?

Up to $29+ million. Pocket change for Disney—and perhaps even Midjourney—but that misses the larger point, which we’ll get to momentarily.

Did I Say That Out Loud?

David Holz, Midjourney's founder, once described his company's approach with startling candor: the model was trained on "a big scrape of the Internet... we weren't picky."

That translate to roughly 100 million images (many still protected by copyright) all fed into the system like coal into a furnace.

To understand why this matters, it will help to better understand how AI training actually works.

Here's a quick, oversimplified rundown of what happens when you train an AI on images:

You feed it millions of pictures: photos, paintings, movie stills, weird art from weird forums, everything. The AI does not save these images, per se. Instead, it notices patterns. Example: It learns that cats usually have pointy ears and whiskers. A crowns on a head suggests royalty (or perhaps the Birthday Boy over at Burger King). That sunsets pair warm colors with horizontal light. Etc.

These patterns get encoded as mathematical relationships: millions of tiny adjustments that help the AI understand what things typically look like and how they relate to each other.

So, when you type something along the lines of "purple toothless dragon sipping tea," the AI combines what it learned: dragons have scales and wings, tea involves cups and relaxed poses, purple suggests magic, no teeth means only gums, etc. But of course, way more granular. It generates something “new” by mixing these visual concepts together.

From Disney’s perspective, they argue that they never gave permission for their characters—their intellectual property—to be part of that learning process. But now when someone asks for an "ice princess," the results often look remarkably like Elsa.

What Does Disney Actually Want?

I am no legal or Disney scholar (though I have watched my share of Disney movies and Judge Judy episodes) so I’ve got to rely on the actual experts for this:

The money isn't the main goal here. Instead, Disney wants two things:

First: a complete inventory of every Disney and NBC Universal asset that ended up in Midjourney's training data.

Second: an injunction powerful enough to shut down Midjourney's newest models until they clean house. This includes their upcoming video generator (a tool that could animate, say, Mickey Mouse without permission).

But injunctions aren't automatic. Judges must weigh whether money alone can fix the harm, and whether shutting down the technology serves the public interest.

“Our world-class IP is built on decades of financial investment, creativity and innovation,” Horacio Gutierrez, a senior exec VP and chief legal and compliance officer for Disney, said in a statement. “[…]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. But piracy is piracy, and the fact that it’s done by an AI company does not make it any less infringing.” - Deadline

Explain Yourself, Midjourney!

In all likelihood, Midjourney will argue fair use, the legal doctrine that is supposed to protect certain uses of copyrighted material. Their case: the AI doesn't store actual pixels from Disney movies. Instead, it learns patterns and statistical relationships, then creates something new when users type prompts.

Kind of like a student studying thousands of paintings to understand how shadows work, then painting an original sunset.

But this defense got harder after the Supreme Court's 2023 Warhol decision. The court said that when a new work competes in the same market as the original, potential economic harm can outweigh claims of artistic transformation.

And Disney characters? They're billion-dollar brands with licensing deals to infinity and beyond.

The Bigger Picture

This case isn't happening in isolation. At least three other major lawsuits are moving through courts:

Getty Images sued Stability AI in London, forcing disclosure of a 12-million-image training dataset. A class action by artists in California survived attempts to dismiss it. The New York Times is battling OpenAI over text outputs that “echo” (ahem, plagiarize) their paywalled articles.

Why $30 Million Misses the Point

In 2024, analysts estimate Midjourney earned around $300 million. There are projections of $500 million for 2025. So, although the damages would sting, it wouldn’t topple the company.

But a forced retraining from the ground up? That could wipe out future earnings entirely.

For Disney, the real prize is establishing precedent for licensed data deals. In the way music producers and publishers transformed streaming from a threat (Napster) into a revenue source (Spotify).

Four Possible Outcomes

So…how will this play out? Experts see four scenarios:

The quiet settlement: Midjourney writes a check and installs filters that block Disney characters from outputs.

The targeted injunction: Prompts like "Elsa surfing" suddenly produce blank or just unrecognizable screens.

The clean slate: A judge orders Midjourney to retrain using only licensed or public-domain art.

The AI victory: Courts declare large-scale scraping fair use, sending rights holders back to Congress for new laws.

The Legal Test That Matters

At the center sits an old copyright question: Does the new work merely replace the original, or does it add something genuinely new?

Courts will weigh this against economic reality. If AI-generated images undercut Disney's ability to license Elsa for lunchboxes—or worse, for Disney's own AI products—fair use becomes a much harder sell.

What This Means for You

Perhaps very little…or or perhaps quite a lot:

The next time you type "Darth Vader doing yoga at sunrise" into an image generator, there are big debates about what consitutes theft.

It's a major test as to whether or not feeding copyrighted art to an AI counts as inspiration or theft.

The answer will reshape how machines learn to “see” and what we, the lowly humans, are allowed to ask them to create.

Sources: Court filings in Disney et al. v. Midjourney, Reuters, The Guardian, public interviews with David Holz, U.S. Supreme Court decision in Andy Warhol Foundation v. Goldsmith (2023)