But the style isn't being mimicked. It's akin to
You must be registered to see the links
. Photos/Drawings/etc. are essentially being taken/stolen (not including paid AIs like Midjourney where they have a dedicated database with a collection of images that effectively become stock images.) and bashed together. Hence why most AI art will have a very soft feeling to them. There's a shit ton of blurring happening to make it all work. So, yeah, software like Stable Diffusion is taking from Google Images, DeviantArt (likely prior to having their own), ArtStation, Pinterest, etc. put them all together based on your prompt/image/etc.
If I steal a whole bunch of images and throw them to together in Photoshop, does that make it mine? No. So, why should that be true for a robot? Hence, why copyright isn't applicable to AI generated/assisted art (
You must be registered to see the links
.). Pro-AI Art shills are pointing to this as a counter:
You must be registered to see the links
, but fail to mention that registration isn't proof of copyright, it's near automatic, and in this case the only copyright is the writing and human additions to the Graphic Novel.
(Edit: Turns out, the copyright office for that
You must be registered to see the links
backtracked. There goes that argument.)
So, basically, if I take a crap on the Mona Lisa (which is public domain) and smear it all over it, only my crap is protected by copyright as it was made by me. The same applies to AI art. It's public domain, effectively at least, and thus only things I add over it by my own hands are protected. This is why it's stupid to pay for AI art that doesn't come out of MidJourney or similar software(s). The whole AI space is ethically and morally gray, and the people who both sell and claim it as their original work are just as bad as those plagiarize and steal other works.
I imagine this is far less of a case with AI writing, though. There's inherently more logic in writing than there is in art, meaning training an AI to write a poem comes a lot easier than drawing/painting/manipulating something (which is why most faces in AI art are terrible.), and thus far easier (used loosely, of course, wouldn't even know where to start on building/coding/etc.) to achieve a good result without outright stealing.
I only found this thread now, but there is A LOT of misunderstanding of how this tech works here.
the style isn't being mimicked. It's akin to
You must be registered to see the links
.
That is absolutely NOT the case. For this to be true, the AI would need to have access to each image in the original database, and it does not in any way. Midjourney runs as a Discord bot, so you can't verify this claim of course, but it is based on the same general principle as Stable Diffusion, and this AI
You must be registered to see the links
and it runs offline. Stable Diffusion was trained with a database of 5 BILLION already compressed images, resulting in a dataset that is 240 terabytes large. The model file that contains all the data the AI needs to create images and all it has access to, however, is only 2 gigabytes large in its simplest format (there are alternative sizes that contain double floating point precision data or additional training data as well, but 2GB is what is required for image generation), and every model is the same size no matter how much "additional" data you train into it, because it does NOT contain the actual images, it contains as the name implies a model of how the AI understands the visual information that related to each word, or token as known in this field, that was taught to the AI. There is no "photo manipulation" involved in the inference process used by AI, instead the
You must be registered to see the links
.
Not that this should matter, because as it turns out...
The whole AI space is ethically and morally gray, and the people who both sell and claim it as their original work are just as bad as those plagiarize and steal other works.
(...)
If I steal a whole bunch of images and throw them to together in Photoshop, does that make it mine?
Kinda does, actually. That's called "a collage", and that's a protected form of art. Even other forms of art famously copied almost directly other artist's works, and that is also protected because it is still art to transform the art of others into something new. Andy Warhol famously took
You must be registered to see the links
and
You must be registered to see the links
that is one of the most influential images ever created, and guess what: he didn't have to ask for permission, because it's fair use. To further the irony of this example, Warhol's image was later used as inspiration for
You must be registered to see the links
by a different artist, Richard Pettibone... who also didn't have to request authorization or inform anyone, because IT IS LEGAL under fair use to transform another artist's work into something new.
But that doesn't matter, because that's NOT what AIs like Stable Diffusion are doing. All they do is learn the techniques and visual styles of existing artists, which as it turns out is what EVERY... SINGLE... ARTIST of any field does to learn their trade. There isn't even one artist in existence that didn't learn by copying and mimicking existing artists that inspired them, and that's fine, because you can't copyright a style, and you shouldn't be able to copyright a style, because copyright isn't meant to protect someone's process or ideas, but to protect their WORK. "Well, someone can use the AI to copy someone's work now very easy", you will probably say. Sure, they can do that... and that will be copyright infringement just like if someone copies that work with a pen, with a brush, or with a digital tablet.
And let's be clear, the argument that this isn't art because somehow the US Copyright Office doesn't accept it is meaningless. The USCO doesn't currently accept AI art, and as you pointed out rescinded the application from that comic because it has still not decided who is considered THE CREATOR in the AI art space. That's an ongoing discussion, but an image is an image, and any image created for artistic purpose is art, which is shown by the fact that the USCO accepted the application in the first place. The only argument here is who owns the copyright of an image created through use of an AI. That's not a moral or philosophical discussion from the point of view of the copyright registration entities like the USCO or other around the world, but simply a legislative one that is currently pending for a few years now, as you can see by
You must be registered to see the links
by the US Patent & Trademark Office.
This whole idea that AIs are "stealing from artists" is plain and simple FUD that people keep spreading because they lack the most basic understanding of how the technology works, and it's easier to share ignorant fearmongering than to try to learn about it, and learn how it helps enable more people to create images they have in their heads, because turns out no AI can make images out of a vacuum: every single one of these AIs require someone to design the prompts that will return the images, and if at one hand they allow tons of images to be created quickly, anyone that attempts to do more than make quick tests learns that it takes A LOT of work to get the AI to create anything more specific, and then to get the resulting images and manipulate them to get a really good result that is better than the average crap you can get straight from the machine. Exactly how anyone can apply a bunch of effects on an image on Photoshop, but it takes work to make anything actually good. Exactly how anyone can make a bunch of lines on a piece of paper with some crayons, but it requires learning techniques and working hard to make a good illustration.
You know what we call someone who spends time using a tool to get the right image they want instead of just random crap? There is a word for that: we call them "artists".
Edit: words
