Does it really have to be entire movies when theres a ton of promotional images and memes with similar images?

I think it’s much more likely whatever scraping they used to get the training data snatched a screenshot of the movie some random internet user posted somewhere. (To confirm, I typed “joaquin phoenix joker” into Google and this very image was very high up in the image results) And of course not only this one but many many more too.

Now I’m not saying scraping copyrighted material is morally right either, but I’d doubt they’d just feed an entire movie frame by frame (or randomly spaced screenshots from throughout a movie), especially because it would make generating good labels for each frame very difficult.

WB and Disney would lose, at least without an amendment to copyright law. That in fact just happened in one court case. It was ruled that using a copyrighted work to train AI does not violate that works copyright.

The way it was done if I remember correctly is that someone found out v6 was trained partially with Stockbase images-caption pairs, so they went to Stockbase and found some images and used those exact tags in the prompts.

The image it generated is really widespread

I have that exact same .jpeg stored on my computer and I don’t even know where it came from. I don’t even watch superhero films

Yeah man, Voyager is making millions with the images on the app. It makes me so mad, they Voyager people make you think they are generating content on their own, but in reality is just feeding you unlicensed content from others.

I just remembered a copyrighted image. Oops.

Hey, I bet there were complaints about Google showing image results at some point too! Lol

Wow, voyager app is very nice!

you’re still asking for a character from a video game, which implies copyrighted material. write the same thing in google and take a look at the images. you get what you ask for.

you can’t, obviously, use any image of Mario for anything outside fair use, no matter if AI generated or you got it from the internet.

If you asked me to draw an Italian video game character, I’d draw Mario too. Why can’t an AI make copyrighted character inspired pics as long as they aren’t being sold?

But where is the infringement?

This NYT article includes the same several copyrighted images and they surely haven’t paid any license. It’s obviously fair use in both cases and NYT’s claim that “it might not be fair use” is just ridiculous.

Worse, the NYT also includes exact copies of the images, while the AI ones are just very close to the original. That’s like the difference between uploading a video of yourself playing a Taylor Swift cover and actually uploading one of Taylor Swift’s own music videos to YouTube.

Even worse the NYT intentionally distributed the copyrighted images, while Midjourney did so unintentionally and specifically states it’s a breach of their terms of service. Your account might be banned if you’re caught using these prompts.

Because this proves that the “AI”, at some level, is storing the data of the Joker movie

I don’t think that’s a justified conclusion.

If I watched a movie, and you asked me to reproduce a simple scene from it, then I could do that if I remembered the character design, angle, framing, etc. None of this would require storing the image, only remembering the visual meaning of it and how to represent that with the tools at my disposal.

If I reproduced it that closely (or even not-nearly-that-closely), then yes, my work would be considered a copyright violation. I would not be able to publish and profit off of it. But that’s on me, not on whoever made the tools I used. The violation is in the result, not the tools.

The problem with these claims is that they are shifting the responsibility for copyright violation off of the people creating the art, and onto the people making the tools used to create the art. I could make the same image in Photoshop; are they going after Adobe, too? Of course not. You can make copyright-violating work in any medium, with any tools. Midjourney is a tool with enough flexibility to create almost any image you can imagine, just like Photoshop.

Does it really matter if it takes a few minutes instead of hours?

I’ve had this discussion before, but that’s not how copyright exceptions work.

Right or wrong (it hasn’t been litigated yet), AI models are being claimed as fair use exceptions to the use of copyrighted material. Similar to other fair uses, the argument goes something like:

“The AI model is simply a digital representation of facts gleamed from the analysis of copyrighted works, and since factual data cannot be copyrighted (e.g. a description of the Mona Lisa vs the painting itself), the model itself is fair use”

I think it’ll boil down to whether the models can be easily used as replacements to the works being claimed, and honestly I think that’ll fail. That the models are quite good at reconstructing common expressions of copyrighted work is novel to the case law, though, and worthy of investigation.

But as someone who thinks ownership of expressions is bullshit anyway, I tend to think copyright is not the right way to go about penalizing or preventing the harm caused by the technology.

Wasn’t that known? Have midjourney ever claimed they didn’t use copyrighted works? There’s also an ongoing argument about the legality of that in general. One recent court case ruled that copyright does not protect a work from being used to train an AI. I’m sure that’s far from the final word on the topic, but it does mean this is a legal grey area at the moment.

Because this proves that the “AI”, at some level, is storing the data of the Joker movie screenshot somewhere inside of its training set.

Is it tho? Honest question.

If the training data is obviously copyright infringement, then the data must be retrained in a more appropriate manner.

This is the crux of the issue, it isn’t obviously copyright infringement. Currently copyright is completely silent on the matter one way or another.

The thing that makes this particularly interesting is that the traditional copyright maximalists, the ones responsible for ballooning copyright durations from its original reasonable limit of 14 years (plus one renewal) to its current absurd duration of 95 years, also stand to benefit greatly from generative works. Instead of the usual full court press we tend to see from the major corporations around anything copyright related we’re instead seeing them take a rather hands off approach.

So let’s say I ask a talented human artist the same thing.

Doesn’t this prove that a human, at some level, is storing the data of the Joker movie screenshot somewhere inside of their memory?

But its not like our laws have changed

And that’s the problem. The internet has drastically reduced the cost of copying information, to the point where entirely new uses like this one are now possible. But those new uses are stifled by copyright law that originates from a time when the only cost was that people with gutenberg presses would be prohibited from printing slightly cheaper books. And there’s no discussion of changing it because the people who benefit from those laws literally are the media.

By that logic I am also storing that image in my dataset, because I know and remember this exact image. I can reproduce it from memory too.

Yes, look how specific they were. I didn’t even need to get that exact with a google image search. I literally searched for “Joaquin Phoenix Joker” and that exact image was the very first result.

They specified that it had to be that specific actor, as that specific character, from that specific movie, and that it had to be a screenshot from a scene in the movie… and they got exactly what they asked for. This isn’t shocking. Shocking would have been if it didn’t produce something nearly identical to that image.

A more interesting result would be what it would spit out if you asked for say “Heath Ledger Joker movie, 2019, screenshot from a movie, movie scene”.

You can hardly consider it compression when you need a compute expensive model with hundreds of gigabytes (if not bigger) to accurately rehydrate it

You can run Stable Diffusion with custom models, variational auto encoders, LoRAs, etc, on an iPhone from 2018. I don’t know what the NYTimes used, but AI image generation is surprisingly cheap once the hard work of creating the models is done. Most SD1.5 model checkpoints are around 2GB in size.

Edit: But yes, the idea of using this as image compression is absurd.

But it’s not deterministic.

deleted by creator

I like that thought too, surely better than calling it AI.

“how much of the data is the original data”?

Even if you could reverse the process perfectly, what you would prove is that something fed into the AI was identical to a copyrighted image. But the image’s license isn’t part of that data. The question is: did the license cover use as training data?

In the case of watermarked images, the answer is clearly no, so then the AI companies have to argue that only tiny parts of any given image come from any given source image, so it still doesn’t violate the license. That’s pretty questionable when waternarks are visible.

In these examples, it’s clear that all parts of the image come directly or indirectly (perhaps some source images were memes based on the original) from the original, so there goes the second line of defence.

The fact that the quality is poor is neither here nor there. You can’t run an image through a filter that adds noise and then say it’s no longer copyrighted.

I purchased a 128 GB flash drive for around 12-15$ (I forgot the exact price) last year, and on Amazon, there are 10 TB hard drives for $100. So, the actual storage doesn’t seem to be an issue.

RAM is expensive 128 GB of RAM on Amazon is $500.

But then again, I am talking about the consumer grade stuff. It might be different for the people who are making AI’s as they might be using the industrial/whatever it’s called grade stuff.

Hmm this sounds like a similar technology to the time cube