Disney sues Midjourney

[Morrus]

That is an odd ruling. To me it seems like if you could train an AI on legally acquired material, you should also be able to do a lot of other things with legally acquired material. AFAIK, copyright law would not allow you to make a movie featuring art on a poster you bought, or text from a book you own, or music from a CD that's yours, and to me this seems like the same thing. Now, I'm of the opinion that you should be able to do those things, at least in an incidental way (just like I don't have to pay Harder & Steenbeck every time I post a photo of a miniature I painted with an airbrush they made), but the law doesn't agree with me.

You are correct, but once the Disneys and the appeal courts etc, have done their inevitable thing, you will no longer be correct. This ruling doesn't stand a chance of lasting a stiff breeze, let alone a Disney-powered hurricane.

 

[Umbran]

From the article: Fair use rulings take into account what the work is being used for (parody and education can be viable), whether it’s being reproduced for commercial gain (you can write “Star Wars” fan fiction, but you can’t sell it), and how transformative a derivative work is from the original.

And it is in that point that Disney will probably win against Midjourney - it sounds like Disney is brining evidence of Midjourney directly reproducing Disney artwork, and effectively advertising their services based on that ability. That, if true, will scuttle the transformative defense.

 

[FrogReaver]

And it is in that point that Disney will probably win against Midjourney - it sounds like Disney is brining evidence of Midjourney directly reproducing Disney artwork, and effectively advertising their services based on that ability. That, if true, will scuttle the transformative defense.

Training the model can be transformative while specific users using the model to produce a copyrighted work could be viewed as infringing. That’s the line I expect to be drawn.

 

[Umbran]

Training the model can be transformative while specific users using the model to produce a copyrighted work could be viewed as infringing. That’s the line I expect to be drawn.

In the past, tools that demonstrably enable large scale infringement have been found to be complicit in that infringement.

 

[FrogReaver]

In the past, tools that demonstrably enable large scale infringement have been found to be complicit in that infringement.

Copy machines and torrents aren’t generally illegal.

 

[briggart]

You are correct, but once the Disneys and the appeal courts etc, have done their inevitable thing, you will no longer be correct. This ruling doesn't stand a chance of lasting a stiff breeze, let alone a Disney-powered hurricane.

I’m not so sure. I feel that this time the House of the Mouse will turn out to be mouse sized after all.

The powers that be seem to consider AI development an area of vital strategic relevance. They won't let something trivial like some tens billions $ worth of copyright infringement get in the way.

 

[Umbran]

Copy machines and torrents aren’t generally illegal.

That's because that's not how copyright enforcement works. It isn't done "generally". It is done in specific.

So, peer-to-peer filesharing technology isn't illegal, but Napster and Kazaa both died under the weight of their lawsuit losses, and the British Government has ordered The Pirate Bay blocked from five internet service providers. And new systems that keep central indexes of content that make large scale infringement easy (as Napster, Kazaa, and Pirate Bay did/do) are rare, due to the legal exposure.

The danger for the industry at the moment isn't that generative AI will become "illegal" - that would take an act of Congress. However, court precedent can make it easy to sue certain implementations into the ground.

I would not be surprised that, after about a decade, we find that "general use" generative AIs, like Midjourney, that can be effectively ordered to reproduce original works, will be easy to sue. Specific use tools that use genAI, but can't easily broadly reproduce originals (like, say, some pattern-fill tools in Photoshop), may end up broadly considered fair use.

 

[Snarf Zagyg]

I’m not so sure. I feel that this time the House of the Mouse will turn out to be mouse sized after all.

The powers that be seem to consider AI development an area of vital strategic relevance. They won't let something trivial like some tens billions $ worth of copyright infringement get in the way.

Very brief background on this (all based on memory, not doing actual sources or checking, so take this FWIW):


1. Why did Disney file this lawsuit, and now (and it's not just Disney- Comcast / Universal / NBC is joining)?

Well, apparently the US Copyright Office was about to release an opinion that stated that LLMs that used copyrighted material was infringement. However, the current administration quashed it (and maybe fired the writer of the opinion? not sure on that). Word is that, like briggart said, certain whispers to the administration from certain people (use your imagination) convinced them that we couldn't allow a mineshaft gap fall behind other countries in AI by allowing copyright claims like this. Disney saw that the usual administrative channels would not work so they went to Court.


2. Why is the lawsuit pled like this?

If I understand it correctly, Disney (and friends) aren't going after the scraping. Instead, they are going after the output. In other words, they are taking the scraping and copying off of the table (for now). Instead, they are making an easier claim- the AI is making infringing derivative works. In other words, regardless of what went in, it outputs infringing works. A lot of them. And that adds up real real quick in terms of damages.


3. Okay, why midjourney?

Because OpenAI has a lot of money. Disney wants to get a good ruling, and it's going after an entity that doesn't have all the money to burn on good attorneys.


This has been a short, and hopefully accurate, summary of what I understand. It's worth what you paid me.

 

[FrogReaver]

That's because that's not how copyright enforcement works. It isn't done "generally". It is done in specific.

So, peer-to-peer filesharing technology isn't illegal, but Napster and Kazaa both died under the weight of their lawsuit losses, and the British Government has ordered The Pirate Bay blocked from five internet service providers. And new systems that keep central indexes of content that make large scale infringement easy (as Napster, Kazaa, and Pirate Bay did/do) are rare, due to the legal exposure.

The danger for the industry at the moment isn't that generative AI will become "illegal" - that would take an act of Congress. However, court precedent can make it easy to sue certain implementations into the ground.

I would not be surprised that, after about a decade, we find that "general use" generative AIs, like Midjourney, that can be effectively ordered to reproduce original works, will be easy to sue. Specific use tools that use genAI, but can't easily broadly reproduce originals (like, say, some pattern-fill tools in Photoshop), may end up broadly considered fair use.

So you agree said tools won’t be illegal.

 

[Umbran]

So you agree said tools won’t be illegal.

I reject your attempt to get me to oversimplify my answer.