Stable Diffusion Frivolous
Some truncated rebuttal excerpts that address common arguments:
We’ve filed a lawsuit challenging Stable Diffusion, a 21st-century collage tool that violates the rights of artists. Because AI needs to be fair & ethical for everyone.
AI art tools do not in any way, shape or form "collage together images". No images are stored in the checkpoints in any way, shape or form, and indeed - as documented below - such a thing is a literal impossibility.
Today, we’re taking another step toward making AI fair & ethical for everyone....[W]e’ve filed a class-action lawsuit against Stability AI, DeviantArt, and Midjourney for their use of Stable Diffusion, a 21st-century collage tool that remixes the copyrighted works of millions of artists whose work was used as training data.
While one can certainly have sympathy for artists who are faced with change in their industry - as has happened many times in the past, to great resistance, such as with the advent of photography,[11] and later, of digital tools like Photoshop[12][13] - the simple facts are, the rights of creators are not unlimited. That's literally what fair use is.[13] In his very critique of AI art "misappropriating" images, the attorney for the plaintiffs takes the images of various researchers straight from their papers, "with no consent" and with "no compensation". And that's fine, because, again, there are limits to the rights of creators, and the world is better for the existence of fair use.
Stable Diffusion contains unauthorized copies of millions—and possibly billions—of copyrighted images. These copies were made without the knowledge or consent of the artists.
There are exceptions (see overtraining / overfitting later), but again, there are NO images stored in AI art tool checkpoints, and they do NOT collage images (which they don't have).
As a lawyer who is also a longtime member of the visual-arts community, it’s an honor to stand up on behalf of fellow artists and continue this vital conversation about how AI will coexist with human culture and creativity.
It should go without saying, but suing someone is not a way to "continue a conversation".[14]
The image-generator companies have made their views clear. Now they can hear from artists.
Note here how the argument is carefully played off as artists vs. AI tool users, without ever feeling the need to actually defend the point. In reality, many of the people using AI art tools were already professional artists.[15] The case takes the anti-progress side of an "artists who fear change" vs. the "artists who embrace change" debate.
---
Worthwhile read. Understand what you hate at least
The idea that people using and making these tools aren't artists is an extremely conservative rebuke to literally the entire last century of art history, so it never ceases to amaze me how blase people are about saying it as though it were the most obvious thing in the world.
This claim extremely questionable:
No images are stored in the checkpoints in any way, shape or form, and indeed - as documented below - such a thing is a literal impossibility.
Here is a well-known painting:
And here is a the first result Stable Diffusion gave me when I used "The Mona Lisa by da Vinci" as a prompt.
And that is a recognizable copy of the Mona Lisa. Clearly quite a lot features of the painting made it into this reproduction, including many details of the figure's clothing and of background objects. It's a terrible copy, but it is a copy, and it shows that somewhere, Stable Diffusion is storing a lot of information about this painting. "Brown sleeves bunched up into many folds" did not just come out of nowhere. "Winding road leading to a lake in the background" isn't a coincidence.
If a computer program can reproduce an image, even in a poor and distorted form, then clearly it stores the information necessary to create the reproduction. And for digital files, there is no difference between "storing information sufficient to reproduce an image" and "storing an image." It may not be stored in a form that's legible to humans, but then a jpeg isn't legible to humans either.
So at best I think you can say "Stable Diffusion stores images, yes, but only with a really lossy and poor quality compression algorithm."
This issue is addressed in the article (pretty comprehensively, imo): the Mona Lisa is a clear example of overtraining--and probably an unavoidable one given that it’s the single most recognizable artwork in the english-speaking world. When working correctly, the model condenses the training images down not to compressed images but to the salient features of those images. The Mona Lisa itself likely shows up thousands of times in the data set, and there are likely thousands more images like this one:
in the dataset as well, to the point where the Mona Lisa itself can be considered a “salient feature” that the model needs to know how to reproduce. This is true for maybe a few dozen works of visual art (Starry Night, American Gothic, Nighthawks, Dogs Playing Poker, etc.) but doesn’t generalize to every image in the model’s dataset. It only holds true for these highly replicated (and highly memed) images. Now, I don’t take the linked article at its word when it says that “nobody wants a subset of the data to be overfit.“ There are absolutely people who want to be able to generate a Mona Lisa with Shrek’s head, or whatever. But it’s not true that Stable Diffusion “stores images” in the general sense, the way that it “stores” this selection of the most-reproduced images in human history.
